{
  "generated": "2026-08-12T08:56:36+05:30",
  "site": {
    "name": "Varun Choraria",
    "bio": "B2B SaaS marketing operator. I decide what marketing should do next, ship the work, and measure it in pipeline or revenue. Consulting, open to senior roles.",
    "description": "B2B SaaS marketing operator. I decide what marketing should do next, ship the work, and measure it in pipeline or revenue. Essays and open-source AI tools.",
    "url": "https://www.varunchoraria.com",
    "writing": "https://www.varunchoraria.com/blog/",
    "topics": ["ai","analytics","career","communication","design","frameworks","growth","leadership","management","product","product-marketing","sales","strategy"],
    "elsewhere": [{"label":"Movie Recommendations","url":"https://letterboxd.com/vcxcvii/films/"},{"label":"The Edge Podcast for CHROs","url":"https://www.youtube.com/@theedgepodcastforchros3212"},{"label":"All talks","url":"https://www.youtube.com/@varunchoraria607/playlists"}]
  },
  "posts": [
    
    {
      "type": "post",
      "title": "My agent found 9.09%. I found one session.",
      "slug": "9-09-percent-of-nobody",
      "url": "https://www.varunchoraria.com/11-sessions-are-not-a-ux-audit/",
      "date": "2026-08-03",
      "tags": ["ai","analytics","design"],
      "excerpt": "My agent found a UX problem on my site. Dead clicks had affected 9.09% of sessions, above the 5% threshold. It filed a GitHub issue and told me what to inspect.",
      "content": "My agent found a UX problem on my site. Dead clicks had affected 9.09% of sessions, above the 5% threshold. It filed a GitHub issue and told me what to inspect.\n\nThere had been 11 sessions.\n\n9.09% was 1 session.\n\nThat was the real bug. The maths was right. The confidence was not.\n\nWhat the agent actually knew\n\nI built a small agent called Lazarus Pit because I knew I was not going to remember to open Microsoft Clarity every Monday.\n\nIt reads the site’s Clarity summary, checks the numbers against a few thresholds, then files a GitHub issue with a possible cause and a suggested fix. It can propose work. It cannot change the site.\n\nOn its first live run, it saw that dead clicks had crossed 5%. The issue recommended checking anything styled like a link or button.\n\nReasonable advice, except the data did not tell the agent which page had the dead click. It did not identify an element. It did not have a recording to inspect. All it knew was that 1 dead click had happened somewhere across the site.\n\nIt had found something worth watching. It wrote it up as something worth fixing.\n\nWhy 9.09% fooled the system\n\n1 of 11 sessions sounds weak because it is weak. 9.09% of sessions, threshold 5% sounds like a finding.\n\nBoth statements are true. Only one makes the size of the evidence obvious.\n\nThe percentage also triggered the workflow. Once the number crossed the threshold, the agent moved straight from observation to explanation to task. Nobody, including me, had made it stop and ask whether the data could support that jump.\n\nThis is not really a sample-size problem. More sessions would make the rate less ridiculous, but a site-wide rate still cannot tell me which button to fix. I had asked one number to answer two different questions:\n\n\n  Did something unusual happen?\n  Do we know enough to change the site?\n\n\nThe agent could answer the first. It pretended it could answer the second.\n\nWhat I changed\n\nI added a floor of 20 human sessions. Below that, Lazarus Pit skips UX findings. Twenty is not suddenly enough evidence. It is just a guardrail against the worst version of the mistake.\n\nThe bigger fix is in how the agent reports what it sees:\n\n\n  Show the count with every rate: 1/11 sessions (9.09%).\n  Keep site-wide data as a site-wide observation.\n  Do not recommend a page-level fix without page-level evidence.\n  Treat one weak window as a note, not a task.\n\n\nI still want the system. The analytics gets read, possible problems arrive where I already work, and nothing changes before I look at it.\n\nBut I had taught the agent what number counts as bad. I had not taught it what evidence counts as enough.\n\nThose are not the same rule."
    },
    
    {
      "type": "post",
      "title": "The models were fine. I was the bug.",
      "slug": "the-models-were-fine-i-was-the-bug",
      "url": "https://www.varunchoraria.com/the-models-were-fine-i-was-the-bug/",
      "date": "2026-08-02",
      "tags": ["ai","management","frameworks"],
      "excerpt": "I typed a command and nothing happened. No output, no error, no prompt coming back, just a cursor sitting there while my laptop got warm.",
      "content": "I typed a command and nothing happened. No output, no error, no prompt coming back, just a cursor sitting there while my laptop got warm.\n\nWhat was actually happening is that my own tool had spawned about 280 copies of itself, each one launching the next.\n\nEverybody writes the post about what their agents did well. Nobody writes this one, which is a shame, because this is where the lessons live. I wrote up what a month of this costs already. Here’s the other column.\n\n280 of me\n\nThe cause is almost funny. When you build a program, the build tool writes out a file. That file needs a little flag on it that says “you’re allowed to run this.” Mine wasn’t setting it.\n\nHere’s the bit worth remembering. A file without that flag isn’t treated as broken. It’s treated as invisible. The shell goes looking for a command by that name, finds a file it isn’t allowed to run, and carries on looking as if the file simply weren’t there.\n\nSo it kept looking, found a small backup script further down the list, and ran that instead. That script’s whole job was to say “can’t find the real thing, let me go fetch it.” Fetching it meant looking up the same name again. Which found the same backup script, which fetched it again, and so on without end.\n\n2 things made this horrible to catch. It never happened to anyone else. When you install software properly, the installer sets that flag for you. It only bit me, on my machine, testing my own build. So it looked random, and random is where you lose days.\n\nAnd there was nothing to search for. It just hung. If a command ever hangs on you with no output, count the processes before you sit there waiting:\n\nps aux | grep -c \"[r]ainmaker\"\n\n\nA number that climbs is a loop. That one line would have saved me an afternoon.\n\nWhat I took from it: an intermittent bug is usually shaped like your machine, not like your code. When something breaks only for you, stop rereading the logic and start listing what’s different about you.\n\nThe agent that fired me mid-task\n\n10 minutes into a session, I’m walking an agent through an interview about a business. Halfway through, it decides the tool it’s running has a flaw, and goes off to rewrite the tool.\n\n25,000 tokens and 10 minutes, and nobody asked it to. The annoying part is that the flaw was real. It wasn’t wrong about the code. It was wrong about whose call it was.\n\nI’d written the rule down. It was in the README, in plain English, saying don’t do this.\n\nThe model doesn’t read the README. It reads the file that defines the skill.\n\nWhat I took from it: a rule written where the model doesn’t look isn’t a rule. It’s a note to yourself.\n\nI moved 5 boundaries into the file it actually loads, and wrote a test for each so they can’t quietly rot:\n\n\n  Don’t modify yourself\n  Don’t edit the user’s site while you’re diagnosing it\n  Don’t ship anything on their behalf\n  Don’t hand-write data a program is supposed to generate\n  Don’t estimate a number the tool can just compute\n\n\nThat last one is my favourite. Guessing at a number that’s one command away is the most human thing on the list.\n\nThe bug that made no sound\n\nThis is the worst of the 4, and it produced no error at all.\n\nThe tool is an orchestrator with 26 smaller skills underneath it. The orchestrator runs the conversation and produces the context. Each small skill refuses to run without that context, deliberately, because a skill that guesses at missing context produces confident nonsense.\n\nI shipped 0.3.1 with the orchestrator in it. The installed plugin was pinned at 0.2.1, which had the 26 skills and no orchestrator.\n\nSo every skill did exactly what it was built to do. It checked for context, found none, and politely declined. 26 correct refusals adding up to a product that did nothing.\n\nNothing crashed and nothing logged. It just sat there doing nothing, very politely.\n\nOnce I went looking I found 3 more of the same shape. A setting written into every config file and read by nothing, because the code that picked it only ever looked at the command-line flag. A startup hook that told the assistant to run a shell command, so it ran the command and stopped instead of loading the skill. A pointer file written to one filename when the assistant reads a different one.\n\nSame bug, 4 hats. The design lived in the prose. The plumbing quietly went around it.\n\nWhat I took from it: the design isn’t what you wrote in the doc. It’s whatever the delivery layer actually does. Check the hook text, the filenames, the version pins, the config keys. That’s usually where the gap is.\n\nMy own gate, in my face\n\nThis is the smallest one, and the one I’d defend hardest.\n\nI published a post with a new tag on it. The deploy failed.\n\nMy publishing pipeline checks every post against an approved list of tags before it’ll build. A tag that isn’t on the list stops the deploy. I’d forgotten I wrote that check, so for about a minute I was annoyed at myself.\n\nThen I remembered why it’s there. Tags with no page behind them. Tags that are near-duplicates of each other. A taxonomy that rots because adding a tag is easier than thinking about one. The check turns a slow invisible mess into a fast obvious failure.\n\nWhat I took from it: the guardrail you wrote will eventually block you, and that’s it working. Deleting it in the moment trades 2 minutes of annoyance today for a mess in 6 months.\n\nSame point applies to agents generally. They’ll do exactly what the guardrails allow. Every rule you don’t write is a rule you’ve decided not to have.\n\nSo what actually broke\n\nRead the 4 back and it’s uncomfortable. None of them were the model being stupid: a permission flag, a boundary in the wrong file, a packaging mistake, and a check working perfectly.\n\nThe models were fine, and my plumbing wasn’t.\n\nI think that’s most of what people mean when they say agents are unreliable. They’ve built something where the design lives in a prompt and the behaviour lives everywhere else.\n\nSo the habit that fixed things wasn’t better prompting. It was writing failures down where the machine reads them again.\n\nFor that tool it’s a file called FEEDBACK.md. 15 entries under “Fixed”, each titled as the symptom instead of the fix. Next to it sits SPEC.md, which says in 5 clauses what “done” means, and every open item gets sorted against it. Anything that doesn’t break a clause gets queued instead of argued about. That one rule took my open list from 6 items to 3, not because things got fixed but because things were finally allowed to leave.\n\nOn my machine there’s a memory folder. One file per fact, summary line at the top, cross-linked. 21 of them. Why a build failed. Why a name is frozen. What a constraint actually was, 6 weeks after everyone stopped remembering.\n\nNone of it is clever. It’s a diary the machine reads before it starts work.\n\nAnd then I did it again\n\nWhile fixing the version pin above, I wrote the setup instructions for my agent into a file called AGENTS.md, which is a sensible name that reads well.\n\nThe assistant I actually use loads a file called CLAUDE.md on startup. It has never opened AGENTS.md in its life.\n\nSo I sat down, wrote careful instructions, saved them, and shipped a tool that had never once read them.\n\nThat is the same mistake as the 2nd one up there. The one I’d already found. Already fixed. Already written a rule about.\n\nMonths in, with the rule sitting in my own notes, I did it again with a different filename.\n\nWhich is the honest ending. None of this stops you being wrong. It just makes the second time cheaper than the first, and there is always a second time."
    },
    
    {
      "type": "post",
      "title": "Who do we sell to vs. Who SHOULD we sell to (and how).",
      "slug": "who-do-we-sell-to-vs-who-should-we-sell-to-and-how",
      "url": "https://www.varunchoraria.com/who-do-we-sell-to-vs-who-should-we-sell-to-and-how/",
      "date": "2026-08-01",
      "tags": ["product-marketing","sales","frameworks"],
      "excerpt": "I’ve been a part of both startups and large organizations, and one consistent trend I’ve unfortunately seen is that ICP is considered a one-off exercise versus the continuous development it should be.",
      "content": "I’ve been a part of both startups and large organizations, and one consistent trend I’ve unfortunately seen is that ICP is considered a one-off exercise versus the continuous development it should be.\n\n\n  You don’t update your ICP after 2 quarters of sales wins.\n  You don’t update your ICP after a feature release based on adoption or key-event usage.\n  You don’t update your ICP after critical accounts churn or you suddenly start losing mid-market customers.\n  You don’t update your ICP until it is too late.\n\n\nThat’s the unfortunate state of the union.\n\nUpdating and validating ICP can be tedious:\n\n\n  You have to rewire sales plays and positioning across sales enablement assets.\n  You have to hold sessions with every stakeholder to ensure they are updated.\n  You have to audit channels, adoption and expansion.\n\n\n“Who do we sell to?” is often lazy targeting: scrape a list based on demographics, technographics and firmographics.\n\n“Who SHOULD we sell to?” is often:\n\n\n  where your GTM motion actually works, meaning you are able to reach and distribute in the channels where your customers hang out\n  where your product team is building features these customers will need\n  where the expansion motion, cross-sell and upsell, looks as good as the initial deal\n\n\nThat is the distinction I want to get into.\n\nThe rest of this runs from most evidence to least. Start where you actually are: customers and a CRM full of closed deals, an audience but no clear pattern, or a product and nobody using it yet.\n\nBeing honest: if you have data\n\nIf you are an established business with customers, start with win-loss analysis.\n\nThis is going to come from 3 key sources:\n\n\n  CRM\n  call recordings\n  validating both honestly and openly with sales\n\n\nOften, data lakes like a CRM have won or lost reasons in a dropdown with a cryptic note from the AE. You want to understand the difference between what actually happened and what was plainly stated.\n\nFor both closed-won and closed-lost accounts, I would closely analyze:\n\n\n  How did the customer hear about us?\n  Which touchpoints hit them, how many were there and whom did they reach inside each account?\n  Where did the deal stall, or where did it spend the most time?\n  What were the objections?\n  Who was in the buyer committee?\n  What was the ACV or ARR?\n\n\nThen stretch this across cohorts. You could segment by deal size, region, employee count or target-team size. If you sell a sales solution to sales teams, for example, cohort by team sizes such as 2–10, 10–100, 100–250 and so on.\n\nThe cohorting methodology should follow the account or ICP qualification filters sales already uses.\n\nRemember, the entire art of pipeline or ICP analysis is to listen to things systematically: through a process, within a defined timeframe and rooted in real data. Then wrap it around a monthly, quarterly and yearly cadence.\n\nOne side produces blame, the other produces criteria\n\nThere are 2 sides to any analysis. One produces blame. The other produces criteria that determine how things should be run the next time.\n\nA simple example:\n\n\n  “Why did the plan upsell not work in region X?”\n\n\nversus:\n\n\n  “Should we have targeted the plan upsell in region X in the first place?”\n\n\nThere are also black-swan cases where prospects who would not have been your best fit turn out to be your best-adopting customers. In these cases, I would investigate the triggers:\n\n\n  What triggered the urgency to buy?\n  Was the team structured differently from similar customers?\n  Did a competitor move trigger this?\n\n\nIt is all about pattern recognition and synthesizing the data well. AI makes that a lot easier now. What RevOps used to take a week to analyze can now take minutes and be accessible and queryable by anyone.\n\nMake new ICP-y friends\n\nEverything above is a very outside-in approach. But what about the prospects who:\n\n\n  visit your website and are deanonymized by intent providers\n  regularly read your newsletters and marketing emails\n  regularly attend your events and webinars\n  read your blogs\n  accept free invitations to conferences\n  engage with your ads\n\n\nI would benchmark these people against the findings so far, with the common denominator being who you initially thought your ICP was.\n\nThe Venn diagram is your sweet spot for understanding how your ICP is evolving.\n\nBuild an industry-friends list\n\nI would add these people to an always-on campaign. Whenever someone publishes a new blog, preferably under a senior leader with a proper author page, send a light FYI-style note.\n\nKeep it simple:\n\n\n  Email one is about the blog, the point of view and how it affects them.\n  Email 2 follows up and asks for their thoughts.\n\n\nWhat makes a difference is sending it like a personal sales note, not a designed marketing mailer. More importantly, personally respond to the emails. A CTA like “tell me your honest thoughts, I read every email” establishes a ton of trust only if you actually do it.\n\nIt is not very scalable. You may need another human or a personalized agent to help. Either way, this is what I call industry friends: a rolling list of people you keep adding to from LinkedIn post engagement, website visitors, enriched emails, past events, customers, partners and so on.\n\nIn parallel, make sure they are actually your ICP. If they are, send them a LinkedIn connection request, manually or through a small PhantomBuster sequence a couple of times a month.\n\nThis creates air cover for the future. If there is a last-minute rush for webinar or event signups, I would expect conversion from this list to be higher. Because you have already established a relationship, conversations at in-person events or even on webinars are immediately warmer.\n\nYou can also identify who they are connected to. If you are trying to get into their account and need an introduction, they will often help.\n\nYou can run thought-leader ads to them too.\n\nWhat if you don’t have ICP-y friends?\n\nYour website is your biggest trump card here. Offer a lot of helpful entry points:\n\n\n  free tools\n  virtual events, which can be re-promoted as always-on or on-demand webinars\n  an on-demand demo, either a click-through Storylane demo or a video behind a simple signup\n\n\nThe cherry on top is that you can identify ICP accounts here too and invite them onto podcasts. This is the only reason I think podcasts in B2B really work.\n\nEspecially in the age of AI, when job descriptions are changing and teams and organizations are being remodeled, you have a real chance to do research through a podcast and make it useful for the larger community.\n\nThat same conversation can become video and carousel content on a personal LinkedIn profile anyway. You can build a no-cost content machine by repurposing a bunch of material and emailing it back to your industry-friends list as it grows.\n\nThis whole thing is an entry-exit, always-on campaign to:\n\n\n  identify ICPs\n  build relationships\n  ensure no account entering your pipeline is coming in cold\n\n\nSomeone needs to know someone, period.\n\nBut what if you just have a website, or not even that?\n\nThis is something I’ve been encountering quite a lot lately because everyone is taking new products to market. More often than not, your product IS going to replace someone else’s.\n\nThe question becomes whether you are replacing the whole suite or only one part of the workflow.\n\nLeaving aside all the V1 product questions, such as how well your product integrates with others, the larger questions are about beta users, alpha customers and which channel you pick.\n\nOutbound seems easiest. You scrape a list and put everyone into a sequence. But that is just spraying and praying.\n\nThere are a few other ways to go about this.\n\nNetwork marketing\n\nYou will have to rely on warm introductions, Product Hunt launches and building in public. Founder brand plays an essential role.\n\nIt can be something as simple as talking about use cases, quick product demos, industry shifts and trends that might get attention.\n\nAttach a minimally intrusive, white-glove, PLG-assisted sales experience. Add phone numbers, product docs and user guides. Use in-product banners and modals alongside a weekly educational nurture sequence.\n\nIf you can tether some product telemetry, identify high-usage cohorts and nudge them towards a paid plan or upgrade.\n\nA generous free tier makes sense if you want to go mass-market and bottoms-up. If you are selling to mid-market and enterprise, do not be as generous, but make crossing the chasm easier.\n\nEarly winners do not win because of price. They win because of value. Build and bundle features accordingly so the conversion rate can improve.\n\nFrom this, you can derive 3 types of ICPs:\n\n\n  People who pay now. These are your early adopters.\n  People who use the product well but do not pay. These are your next-in-line customers. You have to identify the friction, which could be workflow, price, migration or something else.\n  People who do not use the product well and do not pay. Do not jump too quickly to disqualify them. Enrich the data first and see if they are close to whom you want to sell to. If they are, get onto one-to-one calls and understand the gaps in the job to be done. If they are not, disqualify them.\n\n\nReach existing users\n\nThe second option is to reach users of an existing product. Use BuiltWith and Clay to get a list of people and their LinkedIn profiles, then DM them.\n\nOne hack is to DM somebody at the mid-level. They often have the time and enthusiasm to reply and trial your product. Leadership generally does neither unless it is a small startup.\n\nMake outbound less lazy\n\nThe last option is outbound. Pull a list of target accounts, enrich them and create specific plays for each.\n\nThe catch is that this should not be your dream list of accounts. It should be accounts confirmed as users of a certain tech stack or an incumbent.\n\nYou can wire a simple Python script using Claude, or build a Claygent, that searches for [prospect_company].[incumbent_company].com. For example, mutiny.salesforce.com.\n\nSome SaaS products give customers a login subdomain when they pay. The agent simply checks whether that domain exists. It is not proof, but it is one more signal and one of many ways to reduce token or subscription costs while qualifying a list.\n\nIf you have the budget, run paid LinkedIn ads to these accounts and see who engages. Based on that, reach out to the relevant people.\n\nMaking a one-to-many ABM campaign feel like one-to-one can happen in this sequence:\n\n\n  Send 2 or 3 email touches across 12 days.\n  Send a personalized LinkedIn invitation with a product microsite or landing page.\n  Track engagement in real time.\n  Target people who engage with an ad leading to a demo page.\n  Add people who do not engage to the industry-friends list or another nurture sequence.\n\n\nI have seen 7 to 8% demo rates from this. That is my own experience across a handful of campaigns, not a benchmark I would ask you to plan against.\n\nRemember, 100% of your pipeline will not come from one place. You are doing a bit of controlled and strategic spray and pray, then seeing which channels and types of prospects convert faster.\n\nThat feedback should keep changing who you SHOULD sell to.\n\nWhere to start on Monday\n\nPull your last 20 closed deals, 10 won and 10 lost. For each one, answer 6 questions: how they heard about you, who was in the room, where the deal stalled, what the objection was, what it was worth, and whether you would take that deal again.\n\nThe last question is the whole exercise. The first 5 produce a report about what happened. The sixth produces criteria for what you do next.\n\nIf you cannot answer it for most of the 20, that is your finding. You do not have an ICP problem yet. You have a listening problem, and it is cheaper to fix."
    },
    
    {
      "type": "post",
      "title": "The Claude bill said $1,492. I paid $20. Here's how.",
      "slug": "how-to-get-1492-out-of-a-20-claude-subscription",
      "url": "https://www.varunchoraria.com/how-to-get-1492-out-of-a-20-claude-subscription/",
      "date": "2026-07-30",
      "tags": ["ai","analytics","frameworks"],
      "excerpt": "I pay $20 a month for Claude. Last week I went digging through my own logs to find out what that actually buys, because I wanted to write an honest post about it and I did not want to guess.",
      "content": "I pay $20 a month for Claude. Last week I went digging through my own logs to find out what that actually buys, because I wanted to write an honest post about it and I did not want to guess.\n\nThe logs are sitting on everyone’s machine. Every session your coding agent runs gets written to a file, with the token counts attached. Mine covers one month, from the 2nd of July to the 2nd of August. 72 sessions. Older ones had already rotated out.\n\nHere is what came back, by model.\n\n\n  \n    \n      Model\n      Written\n      Cache written\n      Cache read\n      If billed per use\n    \n  \n  \n    \n      Opus 5\n      3.59M\n      9.69M\n      746M\n      $523\n    \n    \n      Sonnet 5\n      2.34M\n      10.7M\n      721M\n      $292\n    \n    \n      Opus 4.8\n      2.40M\n      7.70M\n      313M\n      $265\n    \n    \n      Fable 5\n      1.81M\n      4.81M\n      260M\n      $412\n    \n    \n      Total\n      10.1M\n      32.8M\n      2.04B\n      $1,492\n    \n  \n\n\nOne month. $1,492 of inference at list price, on a $20 plan.\n\nI had been telling people I get about $100 of work out of a $20 subscription. I was off by more than an order of magnitude.\n\nThe number that surprised me wasn’t the money\n\nLook at the last 2 columns again. 10 million tokens written. 2 billion read.\n\nThat ratio is 200 to 1, and it changes what you think you’re paying for.\n\nAlmost none of this is a model producing text. Almost all of it is a model reading. Every time an agent picks up a task, it re-reads the project: the spec, the design file, the last 20 commits, and the rules I wrote for it 6 weeks ago and forgot about. 2 billion tokens of re-reading against 10 million tokens of output.\n\nWhich means the thing you’re actually buying is not writing. It’s context. And the way to get more out of the same $20 is to make the context worth re-reading, not to write cleverer prompts.\n\nThe whole bill\n\nSince the point of this post is receipts, here is all of it.\n\n\n  Claude Pro: $20 a month\n  ChatGPT: free tier\n  Codex: rides on what I already have. 1.05 billion input tokens since February, 797 million of them in July alone\n  Firecrawl: free tier, for bulk scraping\n  context.dev: 10,000 credits, gifted by the founder after I told him I was between roles and couldn’t afford a plan\n  Hosting: GitHub Pages, free\n  Domain: about $10 a year\n  Everything else: nothing\n\n\nThat’s it. There is no agency, no freelancer, no seat licence, no analytics tool with a sales call attached.\n\nHow $20 stretches\n\n3 things, and none of them are clever.\n\nPlan with the expensive model, execute with cheap ones. I use a frontier model to think through what needs to happen, then hand the actual execution to a swarm of smaller agents. Thinking is where the money is worth spending. Typing is not.\n\nCompress the output. I run caveman, a skill by Julius Brussee that makes every response drop articles, filler, and hedging while keeping every number, error string, and line of code exact. Julius measures it at about 65% fewer output tokens. I went looking for that number in my own logs and couldn’t produce an honest version of it, because token counts include code and tool calls, which the skill deliberately leaves alone. So take his figure, not mine. What I can tell you is the effect I actually feel: more work fits inside one session before the context fills up. I didn’t write it, I just use it every day, and it’s the cheapest change on this list.\n\nLet the repo do the remembering. This is the big one, and it gets its own section below.\n\nThe machine\n\nI should mention this because it undercuts the usual excuse.\n\nAll of the above runs on a personal M2 MacBook Air. 8GB of memory. A 256GB drive. In 2026 most people would call that machine finished.\n\nNothing runs locally. The models run somewhere else, the site builds in CI, the daily jobs run on GitHub’s runners. The laptop is a terminal. Hardware stopped being the constraint a while ago and most people haven’t noticed.\n\nWhat my own site actually does now\n\nThe concrete version of all this is the site you’re reading.\n\nEvery push to main runs 10 gates before anything ships. It tests the changelog pipeline, then runs it. A model reads the commit history and writes the visitor changelog, in my voice, because the voice is pinned in a file in the repo rather than re-prompted each time. A bot commits that back. Then dependency audit, 2 syntax linters, post validation, 413 lines of SEO, AEO, and design checks, the build, and a post-build pass over the HTML, schema, images, and links. Only then does it deploy.\n\nA separate job runs every morning at 09:47, pulls the latest release from every side quest repo, and updates the changelogs and the contribution graph.\n\nOne line in that config is doing more work than it looks like it is:\n\non:\n  push:\n    branches: [\"main\"]\n    paths-ignore:\n      - \"_data/site_updates.json\"\n\n\nThe bot commits that file. Without paths-ignore, its own commit retriggers the deploy that made it, which commits the file again, forever. One line stands between a nice piece of automation and an infinite loop billed by the minute.\n\nI did not know that in advance. I found out the way everyone finds out.\n\nThe part that isn’t a flex\n\nI’m learning all of this for the first time. So the honest column of this ledger looks like this.\n\nModels hallucinate at me. My instructions are unclear more often than I’d like to admit, and an unclear instruction burns real tokens producing something confidently wrong. I’ve spent whole sessions on the wrong problem because I never stated what “done” looked like. I’ve had an agent go off and rewrite a tool in the middle of a task I’d asked it to do something else entirely.\n\nNone of that shows up in a $1,492 figure. It’s real, and if you’re starting out you should expect it.\n\nI wrote up the 4 worst ones separately, because the failures turned out to be more useful than the wins.\n\nThe thing that compounds\n\nHere’s what I’d keep if I could keep one habit.\n\nEvery mistake gets written down somewhere the agent will read again.\n\nFor my SEO agent, that’s a file called FEEDBACK.md. It has 15 entries under “Fixed”, and each one is titled as the symptom rather than the fix:\n\n\n  A non-executable bin turned every call into a fork bomb\n  The agent rewrote the tool during a user’s audit\n  Provider keys implied consent\n  A connected analytics property that measured nothing looked healthy\n\n\nAlongside it sits SPEC.md, which defines what “done” means in 5 clauses. Every open item gets sorted against that definition. Anything that doesn’t break a clause gets queued rather than argued about. That single rule took my open list from 6 items to 3, not because things got fixed but because things were finally allowed to leave the list.\n\nAnd on my own machine there’s a memory directory: one file per fact, each with a summary line at the top, each linked to the others. 21 files right now. Why a build failed, why a name is frozen, what a client’s constraint actually is.\n\nNone of this is sophisticated. It’s a diary that the machine reads.\n\nBut it’s the difference between paying for the same mistake twice and paying for it once. When 99.5% of your spend goes on re-reading, what gets re-read matters more than anything else.\n\nThe misconception\n\nThe most common thing I hear is that people want to take a course first.\n\nI can’t do that. I can’t sit through long videos and absorb theory. And I think the premise is wrong anyway. People treat working with these models as a science, and to some extent it is. It’s also an art, and the art is a very old one: conversing with clarity.\n\nThere’s no hack underneath it. Use the model as a thought partner. Pick the right model for the job. State the goal properly. And don’t let it be a yes man, because a yes man will happily help you build the wrong thing at great speed.\n\nThis post is the demonstration. I didn’t write it and hand it to a model to polish. I asked one to interview me, to keep asking follow-ups, to push back when I dodged a question, and then to publish it here in my voice, structured so both search engines and AI agents can read it. It pushed back twice. Both times it was right.\n\nDo it once. Automate it the next time. That’s the entire method, and it costs $20."
    },
    
    {
      "type": "post",
      "title": "Hunting Season for the Rest of Us",
      "slug": "hunting-season-for-the-rest-of-us",
      "url": "https://www.varunchoraria.com/hunting-season-for-the-rest-of-us/",
      "date": "2026-07-24",
      "tags": ["career","ai","product-marketing","frameworks"],
      "excerpt": "About a week into prepping for one of these companies, I could drop the category’s favourite piece of jargon into a sentence without flinching. Felt good about it too.",
      "content": "About a week into prepping for one of these companies, I could drop the category’s favourite piece of jargon into a sentence without flinching. Felt good about it too.\n\nThen I tried explaining to myself why anyone would buy it, and I got about 11 seconds in before I ran out of road.\n\nThat’s the thing nobody warns you about with interview prep. You can accumulate a fairly convincing vocabulary in a weekend, and vocabulary is worse than knowing nothing, because knowing nothing makes you ask questions and vocabulary makes you confident. Confidence invites the follow-up. The follow-up is where you die.\n\nI’m between roles right now, interviewing for GTM and product marketing seats, so I have been finding this out repeatedly and in real time. The market is not helping. Every posting gets a thousand applicants, half of them AI-blasted. AI now does the junior version of most marketing jobs, so “I can execute” stopped being a pitch somewhere around last year. Meanwhile companies expect one senior person plus a few agents to cover what a pod of 3 covered in 2022. Floor went up, ceiling went up, and most of us are standing in the gap between them wondering which way to jump.\n\nI spent a decade deliberately engineering luck and even I’ll admit this season is harsh. You can’t do anything about the market. You can do quite a lot about how you show up to the 4 or 5 shots you get.\n\nSo after that little incident I stopped prepping and spent 3 days building a thing instead. This is what it is, how it’s put together, and what it cost, which was nothing.\n\nWhat I actually built\n\nIt’s a static site. 40 HTML pages sitting in a folder on my desktop, no hosting, no build step, no backend. I open index.html and the whole war room is there.\n\nScreenshots below are anonymized prototypes. I’m mid-process with these companies and publicly dissecting your potential employer is an excellent way to stop being mid-process. The blue callouts are notes on why each piece exists.\n\n\n\nHere’s the folder, roughly:\n\ninterview-dossiers/\n  <company>/raw/<date>/     scraped pages, archived by fetch date\n  research/                 sitemaps, rival homepages, brand data\n  research/rivals/<rival>/  one folder per competitor, shared\n  site/                     the 40 pages\n    index.html              hub\n    <company>/              10 pages, one per company\n    rivals/                 8 competitor teardowns, shared\n    kb/                     link out to the learning base\n    assets/style.css        one stylesheet for everything\n\n\n2 things in there took me 2 bad versions to arrive at, and they’re the only structural ideas in the whole system I’d defend.\n\nThe rivals folder is shared. 3 of the companies I’m talking to compete with overlapping sets of vendors. In version 1 I researched each competitor separately inside each company’s folder, which meant writing the same teardown of the same vendor twice with slightly different adjectives. Now a rival gets researched once, lives in one place, and gets linked from every company that needs it. This is the single reason a new company now takes a few hours instead of a day.\n\nLearning is separate from research. There’s a second folder, gtm-kb, that has nothing to do with any specific company. 4 disciplines in it: ABM, content marketing, SEO plus AI search, and interview craft. Each one is an issue tree, then branch docs, then a self-test. The last question on every self-test is an applied artifact that has to land in an actual company dossier, so studying and prepping feed each other instead of competing for the same evening.\n\nThat split came from noticing that company research goes stale in weeks and industry understanding doesn’t. Mixing them meant throwing away good learning every time a process died.\n\nThe version history, since the failures are more useful than the final state:\n\nv1 was markdown files. Which is what the research skill still outputs, and it’s correct for the machine. It is unreadable for a human at volume. I had maybe 60 files and no way to answer “what do I know about this company” without opening 9 of them.\n\nv2 was one giant HTML page per company. Everything on one scroll. Solved the file-hunting problem and created a worse one, which is that a 4,000-word page is a wall and you don’t read walls the night before a round. You skim them and feel prepared, which is the exact failure mode I was trying to fix.\n\nv3 asked a different question. Not “how do I organize this information” but “what job does this page do for me, and when.” That produced 10 pages per company, each named for its job: overview, 101, 201, strategy and people, market, buyers, competition, fit and objections, drills, sources. I open drills the morning of. I open fit when I’m nervous. I never open all 10 at once, which is the point.\n\nSame lesson as designing without AI slop. The model will generate all of this happily. Deciding what belongs on which page is still your job and it is most of the work.\n\n\n\n3 things on the overview page do the real work, and you can build all 3 in a Google Doc in an hour. The rest of the site is convenience.\n\nThe one insight. Not a summary of the company. The single thing you’d fix or build in your first 90 days, stated like you actually believe it. Mine for one company: they sell several products as separate things and the buyer doesn’t want several things, so the first job is the connective narrative. That’s an opinion, and opinions are what these conversations are secretly testing, because judgment is what we’re all paid for now.\n\n5 numbers you can defend. Sourced, dated, hyperlinked. When you drop a number and someone leans in, “that’s from your own exec’s interview last December” ends it.\n\nThe landmine box. Facts that changed recently and could detonate mid-conversation. In one process I found the CEO had quietly changed a few months earlier. The company’s own about page still had the old titles on it. Picture confidently referencing the wrong CEO in round 1. Now picture being the only candidate in the pile who asks a good question about the transition.\n\nThe primers, which are the highest-leverage part\n\n2 per industry, and this is the direct fix for the 11-seconds problem up top.\n\nThe 101 is the industry from zero, written as a story. Why does this thing exist, who pays for it, what forces them to pay, how does the money actually move. Every term shows up when the story needs it and arrives attached to a reason, which is the only way I’ve found to make terms stick.\n\nThe 201 is what makes you sound like you’ve been in the room before: the debates practitioners genuinely argue about, the unit economics, 20 years of history compressed into 4 acts, and a day in the life of the person who signs the cheque.\n\n\n\nHere’s the difference in practice. Vocabulary-level prep gets you: “they’ve shipped an autonomous agent for the part of the workflow humans always did by hand.” The 201 version gets you: that was never a capability problem, it was a unit-economics problem. The buyer always had the upper hand in those transactions. Actually using it cost more in analyst hours than it saved, so for about 30 years nobody bothered. Agents collapse the cost of paying attention, which is what finally makes the leverage worth exercising.\n\nI can hold a 10 minute conversation on the second one. The first one is a sentence I could have gotten from the homepage.\n\nTo build them: ask whichever model you use to teach you the industry as a narrative, from first principles, with analogies, and a “if you remember one thing” line per section. Then a second pass for the debates, the economics, the history, the buyer’s day. Then read both out loud and mark every sentence you couldn’t defend if someone pushed. Those marks are your study list. Mine was embarrassingly long the first time.\n\nThe tools\n\nClaude Code does the research, through a skill I wrote called interview-recon. It chains into Master Shifu when something case-shaped comes up and needs a framework. Firecrawl handles bulk scraping on the free tier. Context.dev covers what Firecrawl can’t reach, which is LinkedIn public pages, bot-protected review sites, sitemaps, brand data. That one has its own story and it’s at the bottom.\n\nPlain HTML for output because it opens instantly, prints as a leave-behind, and will still work in 10 years when whatever note-taking app I’d otherwise have used has been acquired and shut down.\n\nOne rule about models: use the strongest reasoning model you have for the synthesis step and make it cite a URL for every claim. Not out of distrust exactly. It’s that you’ll be repeating these facts out loud in a room where being wrong once costs you the room.\n\nThe skill is open source, same as Master Shifu, because job hunting is hard enough without paywalled prep.\n\ngithub.com/vcxcvii/interview-recon\n\ngit clone https://github.com/vcxcvii/interview-recon.git ~/.claude/skills/interview-recon\n\n\nThen say “prep me for my interview at [company]”. 3 design decisions in it worth stealing even if you never install it. It announces its backend and its credit budget up front and asks before blowing past it, because most research skills prescribe paid tools and then quietly degrade when you don’t have them. It shows you a research plan before it fetches anything, so you stay the editor. And it appends rather than overwrites, so round 2 loads what round 1 learned and refreshes only what went stale.\n\nThe 4 moves that did the damage\n\nEverything in the war room carries a hyperlink and a date. 4 research moves produced most of what’s actually in there.\n\nRead the earnings calls of the public companies in the space. Private companies don’t publish numbers. Their public neighbours do, and those neighbours set the narrative your interviewer is swimming in whether they know it or not. Transcripts are free on Motley Fool and on company IR pages. One hour with one transcript taught me that the whole industry’s AI revenue reporting is a definitional arms race, everyone counting slightly different things and daring each other to object. That came up in a real conversation within the week.\n\nAudit their content like you already own it. Every site has a public sitemap at sitename.com/sitemap.xml. Pull it, classify the posts by buying intent, chart the result. One company I looked at publishes several times more content than its loudest rival and roughly 70% of it is top-of-funnel fluff. That became a chart and a first-90-days recommendation. Nobody expects a candidate to walk in with the content audit already done, which is exactly why it works.\n\n\n\nMine both kinds of reviews. G2 and Gartner Peer Insights tell you what customers think versus what the homepage claims. Employee review sites tell you what you’re walking into. One company in my pipeline sits under 3 out of 5 across several hundred employee reviews, with the same complaint about pay after promotion showing up over and over. That didn’t kill my interest in the role. It set my posture for the conversation about money, which is research doing a job that has nothing to do with answering their questions.\n\nRead everything the interviewer has published. One hiring manager had posted their strategic thesis months before I ever spoke to them. My prep borrowed their language and pushed it one step further than they had taken it. That isn’t flattery, it’s arriving already aligned instead of spending the first 20 minutes establishing that you might be.\n\nThe part I’d keep if I could only keep one\n\nReading prep is not interview prep. Interviews are spoken, adversarial and on a clock, and none of those things are true of a document.\n\nMy written answers were sharp. My spoken answers were about 40% filler until the third rep. There is no way to discover that by reading.\n\nSo before every round I run live mocks with ChatGPT in voice mode. Copy this and adjust:\n\nYou are playing my interviewer for round 2 at [company].\nHere is their public writing so you can capture how they\nthink and what they care about: [paste posts/talks].\n\nRules:\n- Ask one question at a time, like a real interview.\n- Push back on weak answers. Don't let me ramble past 90 seconds.\n- Interrupt me like a busy executive would.\n- After 20 minutes, stop and grade me on: clarity of my opening\n  point of view, structure under pushback, domain fluency,\n  evidence discipline (did I use numbers), and the quality of\n  my questions to you.\n- Score each 1-10 with specific feedback, and compare against\n  my last attempt: [paste last rubric].\n\n\n2 honest caveats. This simulates the archetype, not the person. You’re practising against what someone with that job and those published opinions tends to value and push on, and a revenue-minded CSO does push differently than a brand-minded CMO. And you have to do it out loud. In voice mode. The entire finding is the gap between how good your answer looks written down and how it sounds coming out of your face.\n\n\n\nThe rubric is what turns this from a novelty into training, because reps without feedback are just rehearsing your mistakes with extra steps. Across 3 attempts my filler words went from 14 to 6. More useful than that, the pushback score taught me something no communication listicle ever managed to make stick: agree with whatever is true in the objection before you start defending. The rubric caught me skipping that. Twice.\n\nEvery question I fumble goes into an objection bank with a written answer, so the same punch doesn’t land twice.\n\n\n\nBuild yours with one prompt: “here’s my resume and the JD, list the 10 hardest objections an interviewer would have about me, what each one is really assessing, and draft my answer.” Then rewrite every answer in your own words, because a script you didn’t write is audible from space.\n\nIf you only have one evening, do these 3: one earnings transcript from the biggest public company in your target’s space, one written point of view about the first thing you’d fix, and one voice mock with the rubric above. Let it humble you tonight so the interviewer doesn’t get to on Thursday.\n\nThe bill for all of it, since I know what job-search finances feel like: Firecrawl free tier, about 500 credits. Context.dev, about 350 credits. Claude Code and ChatGPT were subscriptions I already had, and if you have neither, the free tier of any frontier model covers the primers, the objection bank and text-mode mocks. 3 focused days for the system and the first 3 companies. Incremental cash spent, zero.\n\nThe world is kind\n\nNow the part I didn’t plan for and can’t take any credit for.\n\n5 days ago Yahia Bakour, who founded Context.dev, sent me one of those routine founder check-in emails. Saw you signed up, how’s the API treating you. I told him the truth, which was that I loved the product, I was building an interview research agent on top of it, and I couldn’t afford a paid plan right now because I’m between roles.\n\nHe replied the next morning. He’d dropped 10,000 credits into my account, on the house, and bumped my rate limit. No strings, no call, no “let’s find a way to work together.” When I offered to do some free marketing work in return, his entire ask was that if I ever meet someone who needs the API, send them his way.\n\nMy whole thesis about luck is that it compounds on honest effort made visible, and I still believe that. But sometimes it’s much simpler than the thesis. Sometimes a stranger is just kind, at the exact moment when kindness lands hardest. Half the research depth in this post runs on credits I was given by someone with no reason to give them.\n\nSo, holding up my end. If you’re building anything that needs clean web data, scraping, extraction, brand intelligence, sitemaps, go and try context.dev. It’s accurate in the places where other tools fail silently, and it’s built by someone who treats a broke job-seeker like a future customer rather than a freeloader. That’s a founder worth betting on.\n\nWhen this season ends I’ll publish the recap: real screenshots, and an honest account of what actually worked once I was in the rooms. The rest of what I build in public lives on side quests, and if you’d rather hire this thinking than copy it, here’s how I work.\n\nThe market is harsh and nobody is coming to make it less harsh. But the same tools that raised the bar will get you over it, mostly for free, if you prep like an operator instead of a candidate.\n\nHappy hunting. If any of this helps you land something, tell me. That’s the whole reason for writing it down."
    },
    
    {
      "type": "post",
      "title": "Github-ing",
      "slug": "github-is-where-ideas-become-accountable",
      "url": "https://www.varunchoraria.com/github-ing/",
      "date": "2026-07-16",
      "tags": ["ai","strategy","career"],
      "excerpt": "The world is moving toward systems-first work, and I do not think most people understand what that means yet.",
      "content": "The world is moving toward systems-first work, and I do not think most people understand what that means yet.\n\nEvery room has some version of the same conversation.\n\nWhat skills can we build? Which workflows can we connect? Can one agent hand work to another? Can the whole process run without 6 people chasing each other across Slack, email, and a spreadsheet called final_v7?\n\nThese questions used to live with engineers, operations people, and the one person in every company who enjoyed Zapier a little too much.\n\nNow they are table stakes.\n\nThe macro is the point\n\nCapital is tighter. Customer wallets are tighter. Tools appear and disappear faster than procurement can finish reviewing them.\n\nI do not believe every company is about to become a 10-person unicorn. I do believe every leadership team has now seen what unusually small, AI-native teams can produce. You cannot unsee that.\n\nARR per employee used to be finance hygiene. Now it is an operating philosophy.\n\nThe question inside companies has changed from “how do we make everyone more productive?” to “how much meaningful capacity can this system create per employee?”\n\nProductivity is doing the same work faster.\n\nCapacity is being able to take on more useful work without the organization collapsing under its own coordination cost.\n\nThat is a much bigger change.\n\nI wrote earlier about leading a small team with a mandate to become AI-native. The goal was not to generate more documents. It was to increase the amount of directionally correct work we could move from idea to market.\n\nSpeed plus direction. The physical definition of velocity, applied to an organization.\n\nThis is also why I think a lot of middle management is in trouble.\n\nNot all of it. Judgment, coaching, conflict resolution, and decision-making remain real work.\n\nBut a lot of management is information routing dressed up as leadership. Collect status. Reformat status. Present status. Send status back down with comments.\n\nAI will attack that layer before it replaces good judgment.\n\nOrganizations will get flatter because the cost of coordination has become visible, and because machines are getting good at the administrative work that kept many layers busy.\n\nThat is the macro. GitHub is one of the consequences.\n\nGitHub is where ideas become accountable\n\nPut bluntly, GitHub is where ideas become accountable.\n\nAI has made producing things dramatically cheaper. A nontechnical person can now build a calculator, draft a workflow, create a landing page, analyze a dataset, document a process, and deploy a working tool.\n\nI still do not understand code beyond basic HTML.\n\nBut software has always had 2 different forms of literacy.\n\nOne is understanding code.\n\nThe other is understanding how software works: inputs, outputs, dependencies, failure states, users, and whether the thing creates an outcome anybody values.\n\nAI has lowered the cost of the first one. It has not removed the need for the second.\n\nThe cost of producing code is collapsing. The cost of deciding what should exist is not.\n\nGitHub is the modern Google Drive for people like me, except it does not have to become a graveyard of ideas that could have been.\n\nA repository can show the work changing.\n\nIt can show what was added, what broke, what got removed, who decided, and why the current version exists.\n\nThat makes GitHub less like storage and more like connective tissue across tools, people, products, and decisions.\n\nGitHub is not your database\n\nThere is a reasonable argument against putting everything in GitHub.\n\nCursor built ChatGTM, an internal system that pulls live information across accounts, product usage, and CRM data, then synthesizes it for teams.\n\nHightouch described another model, with GitHub acting as a system of record for documentation, logic, rules, and skills.\n\nThese can look like competing choices.\n\nThey are not.\n\nLive customer and operational data should live in systems designed for live data. Your CRM, warehouse, product database, and analytics layer exist for a reason.\n\nGitHub is better for the durable logic around those systems.\n\nDefinitions. Documentation. Agent skills. Workflow rules. Prompts that became infrastructure. Configuration. Decision records. The code connecting one system to another.\n\nThe live number belongs in the database.\n\nThe explanation of what the number means, who owns it, how it is calculated, and what should happen when it changes belongs somewhere versioned.\n\nThat distinction protects accuracy without making the company dependent on whichever AI tool is fashionable this quarter.\n\nDocumentation is part of the product\n\nMy opinion is that documentation itself is becoming a product.\n\nNot documentation as a pile of meeting notes.\n\nDocumentation as the operating logic of the business.\n\nEvery organization is broken in its own way. Reporting lines are different. Dotted lines are different. Political constraints are different. Data quality is different. Even the definition of a qualified lead can trigger a minor civil war.\n\nThat uniqueness is exactly why one generic AI tool cannot become the operating system for every company.\n\nThe tools will change.\n\nThe durable layer is the business, technical, and economic logic describing how your company works.\n\nVersioning changes how that logic can be managed.\n\nYou can review why a workflow changed. You can compare the old definition with the new one. You can see when a rule created an unintended consequence. You can schedule an agent to find duplicates, contradictions, dead steps, and instructions nobody follows.\n\nMore importantly, you can reverse a bad decision without reconstructing history from somebody’s memory.\n\nThat is not developer hygiene.\n\nThat is organizational memory.\n\nGitHub is also a portfolio\n\nPortfolios used to belong to designers and engineers.\n\nThat excuse is gone.\n\nThe barrier to building a useful lead magnet, internal tool, research workflow, or small product used to be high enough that a marketer could reasonably say, “I had the idea, but I needed engineering.”\n\nNow many of these things can be built with a frontier model in an afternoon.\n\nNot perfectly. Not always safely. Not without judgment.\n\nBut built.\n\nWould I refuse to hire a marketing or GTM person because they do not have a GitHub profile? No.\n\nWould the burden of proof be higher than it used to be? Absolutely.\n\nA resume tells me what somebody was near.\n\nA repository can tell me what they thought, what they attempted, what failed, and whether they returned to improve it.\n\nThat is much closer to proof of work.\n\nI wrote about this before: do the work once, automate it the next time, then document what happened. Over time, that becomes a portfolio without needing permission from a job title.\n\nThe old line says ideas are cheap and execution is everything.\n\nIdeas are now cheaper than ever.\n\nThe valuable part is the trail between the idea and the outcome.\n\nYou do not need to become an engineer\n\nThis is not an argument that every person should learn computer science.\n\nIt is an argument that every knowledge worker should understand systems well enough to make their work inspectable.\n\nLearn enough GitHub to create a repository.\n\nLearn enough Markdown to document a decision.\n\nLearn enough version control to see what changed.\n\nLearn enough about software to know when the model is confidently wrong.\n\nThen build something useful and leave a trail.\n\nAI gave almost everyone execution leverage.\n\nGitHub gives that execution memory, evidence, and accountability.\n\nThat is why it matters."
    },
    
    {
      "type": "post",
      "title": "How to design without AI Slop",
      "slug": "how-to-design-without-ai-slop",
      "url": "https://www.varunchoraria.com/how-to-design-without-ai-slop/",
      "date": "2026-07-08",
      "tags": ["ai","design"],
      "excerpt": "You are what you admire",
      "content": "You are what you admire\n\nIf you’re tired of AI spitting out the same generic slop every time you ask it to design a landing page, the fix is simpler than it looks. It starts with taste.\n\nAs a marketer, good taste is close to a mandate. Marketing is a horizontal function: you’re stitching design, copy, and math into a single campaign, and it all has to feel like one thing.\n\nAlmost 10 years ago I kept a swipe file. Vintage ads, headlines I loved, layouts I’d happily steal from. I’d flip through it before starting anything new. That swipe file was the prehistoric ancestor of the taste.md file doing the rounds on X right now.\n\nA taste.md distills what good looks like: font pairings, color, WCAG contrast, spacing, the small decisions that separate considered from generic. You can write your own if you’re a designer, or, like me, reverse-engineer designs you admire and build on top of them.\n\nThat’s the whole trick. You are what you admire, and so is an LLM. Feed it slop references and it hands you slop. Feed it taste and it finally has something better to imitate.\n\nWalking the talk\n\nI redesigned this site this week (here’s the changelog). The old version was a busy, sidebar-heavy blog with a blue gradient and a nudge toast. I wanted something quieter and more modern. Before, and after:\n\n\n\n\n\nHere’s the exact process, and one you can copy.\n\nSpin up Claude and switch to the best model. If you can’t stretch to Claude Pro, or you’d rather run something like opencode, point it at my repo, michealangelo. It’s a skill pack you install alongside your agent, not a replacement for it. It encodes UI/UX principles, web guidelines, and design psychology into guardrails, plus a design and animation vocabulary so you can give precise feedback (“the easing feels mechanical”) instead of vague feedback (“make it nicer”).\n\n\n\nTreat it like a senior designer, not a pair of hands. A senior designer doesn’t push pixels all day. They exercise taste and judgement and direct the execution. Talk to the model that way.\n\nFeed it what you admire. Drop in screenshots of sites you love, palettes, fonts. Ask it to build on top, then add your own flavor. I wanted shadcn components for consistency and hierarchy. They’re free and genuinely handy.\n\nGet a written brief before anyone touches code. Ask Claude to generate a detailed design brief for whichever model will execute (I targeted Codex, then Claude Code). Annotating a live preview burns tokens fast. A brief is cheaper, and it travels between tools.\n\n\n\nExecute in the CLI. I use Claude Code partly because caveman ultra cuts token spend by roughly 65% (measured) per session. Pick a mid-tier model on medium effort to execute, and save the frontier model for the thinking. Ship it for free on GitHub Pages, or on Vercel.\n\n\n\nWatch your spend. Something like CodexBar keeps token usage visible. This whole update, adding an “ask any AI about me” band and restoring my RSS feed to the nav, cost about 19% of a single session.\n\n\n\nFor feedback, render locally. Ask the agent to serve it on localhost. Want to compare versions? Spin up a few ports at once (I ran 4001, 4002, and 4003 side by side: ancient, colored, and the new monochrome build). Giving feedback is easy: just paste screenshots back.\n\n\n\nThe part everyone skips: enforce it\n\nHere’s what actually kills slop, and what most “just use a taste.md” advice leaves out.\n\nA taste.md is a suggestion. Suggestions rot. 3 sessions in, an agent quietly reintroduces a stray accent color or swaps your font on a button, and you don’t notice until it’s live.\n\nSo I didn’t just document taste. I made it a rule the machine checks.\n\nMy site is now strictly monochrome: one zinc scale, and the only color left is in photographs. That decision lives in a machine-readable DESIGN.md. And a pre-push QA gate fails any commit that reintroduces a hue, a legacy class, or the wrong font on UI. If an agent, or me at 1am, tries to sneak color back in, the push is blocked before it ships.\n\nThat’s the shift: taste as a test, not a mood board. The mood board tells the model what good looks like. The test guarantees the codebase stays that way, no matter how many agents touch it after you.\n\nSlop isn’t only a generation problem. It’s an entropy problem. Guardrails set the direction; enforcement holds the line.\n\nSo\n\nYou are what you admire, so curate what you admire, then encode it. Write the swipe file. Turn it into guardrails. Make the guardrails enforceable. Do that and the generic look doesn’t just leave, it can’t come back."
    },
    
    {
      "type": "post",
      "title": "I now have my own Master Shifu",
      "slug": "i-now-have-my-own-master-shifu",
      "url": "https://www.varunchoraria.com/i-now-have-my-own-master-shifu/",
      "date": "2026-07-02",
      "tags": ["ai","frameworks","strategy"],
      "excerpt": "I’ve been collecting consulting casebooks for about a decade. Kellogg, Wharton, Ross, Darden, IIM (A, B, C, I, L), MDI, Drucker, McKinsey, Victor Cheng, Crack the Case — 19 books in total, some of them 300+ pages.",
      "content": "I’ve been collecting consulting casebooks for about a decade. Kellogg, Wharton, Ross, Darden, IIM (A, B, C, I, L), MDI, Drucker, McKinsey, Victor Cheng, Crack the Case — 19 books in total, some of them 300+ pages.\n\nThey teach you how to think about business problems. Profitability, market entry, pricing, M&A, operations — each framework is a lens. A structured way to decompose something messy into something solvable.\n\nFor years they sat in my iCloud collecting dust. I’d pull one out before a big work presentation or to prep for a strategy conversation, but the friction of finding the right framework, remembering the steps, applying it correctly under pressure meant I mostly just winged it.\n\nLike most people do.\n\nLast month, I fed every single book to Claude and turned each framework into a /command. Named the whole thing Master Shifu — because every AI agent needs a cranky old panda who forces it to think before it acts.\n\n\n\nWhy frameworks fail in practice\n\nHere’s the thing about frameworks: they’re not the problem. The problem is how we use them.\n\nWalk into any meeting where someone says “let’s run a SWOT analysis” and watch what happens. People brainstorm strengths for 5 minutes, skip weaknesses because it feels negative, make up a couple threats, and call it a day. The framework was present. The rigor was absent.\n\nI’ve done this. You’ve done this. We’ve all done this.\n\nA framework is only as good as the discipline with which it’s applied. And discipline is scarce. Especially when you’re in a room with 6 other people, the clock is ticking, and the CEO is looking at you expectantly.\n\nThat’s where the idea started. Not “let’s build an AI that knows frameworks” — there’s a million of those. But “let’s build an AI that doesn’t skip steps.”\n\n\n\nThe 5 phases\n\nThe flow is simple but the difference is in the order.\n\nWhen you tell Master Shifu a problem — say, “my SaaS margins are shrinking and I’m thinking about entering Brazil” — it doesn’t jump to conclusions. It runs 5 phases:\n\n1. Grilling. Round after round of clarifying questions. “What data do you have? How do you define margin? Is revenue growing or flat? What’s your current unit economics? Do you have any competitors in Brazil already?” It’s annoying on purpose. Most problems fall apart in the first 5 minutes if you ask the right questions.\n\n2. 5 lenses. Before picking a framework, it looks at the problem from 5 angles: customer, competition, financial, operational, and market. Each lens can auto-populate with real data via web scraping. The competition lens will check if your competitor changed pricing last week. The financial lens can pull SEC filings. The market lens scrapes industry news.\n\n3. Framework match. Based on the enriched context, it selects 1-3 frameworks. Not a random guess — pattern-matched against the problem language and the lens outputs. It might return profitability + industry analysis + pricing. Or market entry + Porter’s Five Forces + break-even. Depends on what the problem actually needs.\n\n4. Apply. Each framework runs end-to-end. Step by step. It doesn’t skip the unattractive quadrant of Porter’s Five Forces. It doesn’t forget to calculate contribution margin in a break-even analysis. It applies the same rigor every time.\n\n5. Synthesize. The outputs from multiple frameworks are merged into a coherent recommendation. If profitability analysis pinpoints the problem as volume decline in Europe, and industry analysis flags a new entrant with lower pricing, the synthesis connects those dots explicitly.\n\nI’ve caught myself reading the output and thinking “huh, I wouldn’t have made that connection.” That’s the whole point. I’m not outsourcing my thinking — I’m using the AI to catch the gaps in it.\n\n\n\n32 frameworks\n\nCurrently there are 32. The heavy hitters:\n\n\n  /profitability — P = R - C, segmented by product/channel/region\n  /market-entry — attractiveness → feasibility → competitive response\n  /pricing — cost-based, competitor-based, value-based with elasticity\n  /industry-analysis — Porter’s Five Forces, soup to nuts\n  /m-and-a-fit — synergies, one-way exploitation, integration risk\n  /issue-tree — MECE decomposition, hypothesis driven\n  /growth-strategy — Ansoff Matrix, the whole grid\n  /bcg-matrix — stars, cash cows, question marks, dogs\n\n\nPlus the less glamorous ones that matter more than people admit: due diligence, non-profit strategy, capacity expansion, outsourcing decisions, supply-demand analysis.\n\nBecause the real world isn’t a case interview. Sometimes you need to figure out whether to build a second factory in Vietnam. That’s a capacity expansion problem, not a profitability problem, and the tool should know the difference.\n\n\n\nThe meta-skill twist\n\nThere’s a skill called task-observer that watches every session and logs what the agent got wrong.\n\nAt the end of a heavy session, I just ask “Any observations?” and it surfaces things like:\n\n\n  “You ran the profitability framework but didn’t segment by channel. The issue tree would have caught this. Update the profitability skill to cross-reference issue tree for root cause isolation?”\n\n\nOr:\n\n\n  “The market entry analysis used outdated competitor data. Schedule the competition monitor to refresh weekly.”\n\n\nMeta, I know. But it’s the closest thing I’ve built to a self-improving system. Every session makes the next one better.\n\n\n\nThe deeper point\n\nI’ve been thinking a lot about what happens when your AI is no longer a question-answer machine but an actual extension of how you reason about problems.\n\nThe ceiling isn’t on the framework side — anyone can google “Porter’s Five Forces” and get a template. The ceiling is on the application side. Running it without bias, without skipping the hard parts, without rushing to confirm what you already believe.\n\nThat’s what Master Shifu does for me. Not because the AI is smart — because the structure is enforced. The scaffolding holds.\n\n\n\nIt’s open source, MIT licensed. Because structured thinking shouldn’t be paywalled.\n\ncurl -fsSL https://raw.githubusercontent.com/vcxcvii/master-shifu/main/install.sh | bash\n\n\nOr on my GitHub: github.com/vcxcvii/master-shifu\n\nIf you’re a consultant, founder, PM, or just someone who wants their AI to think in frameworks instead of vibes — it’s yours. Steal it, fork it, break it, improve it. That’s what open source is for."
    },
    
    {
      "type": "post",
      "title": "Nobody uses my MCP server. Building it still worked.",
      "slug": "nobody-uses-my-mcp-server-building-it-still-worked",
      "url": "https://www.varunchoraria.com/nobody-uses-my-mcp-server-building-it-still-worked/",
      "date": "2026-06-28",
      "tags": ["ai","career","strategy"],
      "excerpt": "Nobody’s going to read this, and I don’t mean this post specifically. I mean the whole thing: the blog, the about page, the work history. A hiring manager has 400 applications open and 11 minutes. A reader lands on one essay from a link and leaves. Resumes are boring by design, because the format won’t let them be anything else.",
      "content": "Nobody’s going to read this, and I don’t mean this post specifically. I mean the whole thing: the blog, the about page, the work history. A hiring manager has 400 applications open and 11 minutes. A reader lands on one essay from a link and leaves. Resumes are boring by design, because the format won’t let them be anything else.\n\nWho deep dives into a stranger’s rabbit hole in 2026?\n\nSo on 28 June I stopped trying to get people to read me, and shipped something their AI could read instead.\n\nWhat it actually is\n\nAn MCP server sitting at a URL. You plug it into Claude, or ChatGPT, or whatever coding agent you use, and it can then read this site directly. Every post, every page, the tags, the dates, the full text.\n\nThe point isn’t the protocol. The point is what it lets someone do without reading anything.\n\nA hiring manager can ask “has he actually shipped product marketing, or does he just write about it” and get an answer built from the whole record instead of from whichever page they happened to open.\n\nA reader who liked one post can ask what else I’ve written on that topic, and, more usefully, how the thinking changed between them. That’s the thing a blog is genuinely bad at. Posts are filed by date. Nobody reads a blog chronologically. The evolution is the interesting part and it’s the part the format hides.\n\nAnd agents can use it, which is the same reason there’s an llms.txt sitting there.\n\nLaunch day went great\n\nHere’s the commit history from 28 June, in order.\n\nAdd /mcp/ and /feed/ pages. Update MCP page, HTTP server at the Vercel URL. Fix MCP server URL. Redesign MCP page: hero URL, copy buttons, 6-tab layout, mobile-first. Fix tab and copy buttons, add why-MCP content, rewrite prompts. Add MCP in-action screenshots. Reorder page, shrink demo images, fix caption copy.\n\nThen the next morning: remove broken Connect-by-client tabs, fix mobile layout.\n\nSo I shipped a page with the wrong server URL on it, then built 6 tabs of connection instructions, then discovered the tabs were broken and deleted them the following day.\n\nA month later I found another one. The MCP feed was serving unreadable page slugs and listing pages with nothing in them. Which means for roughly 4 weeks, anything that did connect got a partly broken index of my site.\n\nNobody told me. I found it myself, by accident.\n\n5 weeks later, the logs\n\nHere’s what the server has actually done in the last 30 days.\n\n\n  5,721 requests\n  5,713 of them returned 200\n  Every single one a GET /\n\n\n191 requests a day. If you stop there it looks like a success, and for about a minute I let it.\n\nThen I looked at the raw lines instead of the total.\n\nThey arrive in bursts. 1 to 2 requests per second, back to back, every one a cache miss. That’s not a person asking a question. That’s not an AI client either. That’s a monitor, or a crawler, or something on a loop.\n\nAnd they’re all GET /. MCP tool calls are JSON-RPC, so a GET to the root returning 200 is a handshake or a health check. Something is knocking. Nothing is coming in.\n\nI can’t prove none of those 5,721 were real. I can say that the shape of the traffic looks nothing like use.\n\nAnd zero people\n\nIn 5 weeks, not one person has told me they used it. Not a hiring manager in an interview, not a recruiter, not a reader replying to a post.\n\nWhat has happened, a few times, is that people notice it exists. Someone clocks that there’s an MCP server on a personal site and says something about that. Not about anything it told them.\n\nWhich took me a while to sit with, because it’s a smaller thing than I wanted. I built a channel. What I got was a signal.\n\nThe bit worth keeping\n\nThat distinction matters more than the traffic number, so here it is plainly.\n\nA channel is a way for people to get something from you. A signal is what building the thing tells people about you.\n\nI aimed at the channel. I hit the signal. And the signal is doing real work: it says I ship, it says I understood MCP early enough to have an opinion, and it says the claim on my consulting page about being AI-native has something behind it other than the claim.\n\nNone of that requires a single person to run a query.\n\nWould I have built it if I’d known nobody would use it? Yes, and I want to be honest about why, because “it was strategic” would be a lie invented afterwards.\n\nAt the time I put it at 50-50. Genuinely. Half of me thought MCP becomes a normal way people meet a person or a company, the way an RSS feed once was and a LinkedIn page is now. The other half thought it’d be a curiosity nobody touches.\n\nI built it on a coin flip because the downside was a weekend and the upside was being early.\n\nThat’s the actual calculation, and it’s the one I’d defend. Not “this will be a channel,” which I didn’t know. “The cost of being wrong here is one weekend.”\n\nWhat I’m doing next\n\nSame bet, bigger version. I want a voice clone that answers as me, and a real chatbot built on everything I know rather than everything I’ve published. The server was step one, and step one taught me the interesting problem isn’t access. Access is solved. It’s that a static index of posts doesn’t sound like a person, and sounding like a person is what makes anyone bother.\n\nIf you want to test the version that exists, it’s here. Connect it and ask it something I’d struggle to answer well about myself.\n\nAnd if you do, tell me. You’d be the first."
    },
    
    {
      "type": "post",
      "title": "Maybe going around CIRCLES is worth it",
      "slug": "maybe-going-around-circles-is-worth-it",
      "url": "https://www.varunchoraria.com/maybe-going-around-circles-is-worth-it/",
      "date": "2026-06-26",
      "tags": ["product","analytics","frameworks","growth"],
      "excerpt": "Frameworks get a bad rep. But early in your problem-solving journey, they’re scaffolding — not a crutch. Over time, as you get acquainted with more real-life scenarios, your intuition gets powerful enough to spot-check and work bottoms-up.",
      "content": "Frameworks get a bad rep. But early in your problem-solving journey, they’re scaffolding — not a crutch. Over time, as you get acquainted with more real-life scenarios, your intuition gets powerful enough to spot-check and work bottoms-up.\n\nFor the purpose of this field note, let’s stick to some 101s.\n\nCIRCLES is a helpful framework to bring a rounded thinking hat to the table and actually ask good, focused questions. I first read this in cracking the PM interview.\n\n\n  \n    \n       \n      Step\n      Description\n    \n  \n  \n    \n      C\n      comprehend the situation\n      clarify what’s within scope and out of scope for the problem statement. eg: business goal, constraints\n    \n    \n      I\n      identify the customer\n      figure out how to segment the customers. low hanging fruit will always be the biggest customer segment with the largest gap. list down firmographics, demographics, technographics - get your fact sheet in place about the customer’s day in the life.\n    \n    \n      R\n      report the customer’s needs\n      list all their needs (not wants) in the order of priority. I wrote a helpful mental model here.\n    \n    \n      C\n      cut through prioritization\n      narrow the problem surface. from the prioritized needs, pick the highest-impact, most-urgent ones to address — don’t try to solve everything at once.\n    \n    \n      L\n      list down solutions\n      generate 3-5 solutions per prioritized need\n    \n    \n      E\n      evaluate tradeoffs\n      this needs to happen across 3 axes - implementation complexity, user impact, cost / benefit\n    \n    \n      S\n      summarize\n      this is where you need to COMMIT. it’s fair to be opinionated here, because you need to pick a solution and defend it. In the age of agentic AI, we’re all compensated for context and judgement.\n    \n  \n\n\n\n\n\nThe same thing works in RCAs\n\n\n  \n    A costly lesson I learnt in my career: check the damn dashboard itself for telemetry errors, sanity of the data lake and how a metric is computed. You’d be surprised how often that occurs. Plumbing issues like hygiene and data pipeline breakages often misreport signals.\n  \n  \n    Second, check for seasonality. In B2C, holiday season has an impact on business. In B2B, it’s a factor of sales cycle length, targeting the right buying committee, competitor takedowns etc (it is complex).\n  \n  \n    Another dumb error I’ve made in the past is not to check for things that might’ve gone wrong on our end in production. Was there a server downtime? Did API response times fall? Was the customer’s plan not grandfathered after a price change?\n  \n  \n    Segment the surface area that was affected: check traffic sources, user segments like mobile vs. desktop, feature adoption data, past usage history, past support tickets etc.\n  \n  \n    Most of the above 4 pointers are internal-facing. Externally - you can look at things like competitor launches, industry trend causing the dip, customer awareness about your product / features (enablement).\n  \n\n\nOther helpful things you can do:\n\n\n  \n    Setup observability. There’s options like new relic (freemium); signoz (open-source). Or, you can also vibe-code a dashboard or even build one on google sheets and have that updated real-time via a tool like coefficient. You can then hook it up to an LLM and run a /routine or cloud agent where it pings you daily updates on slack.\n  \n  \n    For the love of god, track conversion funnels on both google analytics as well as your internal product analytics platform like heap.\n  \n  \n    You can also setup an RSS or a custom competitor scraping agent via firecrawl or context.dev to scrape product pages, newsroom / PR, product hunt, app exchange listings etc via an LLM and connect it to slack for real-time updates. Costs super less.\n  \n\n\nI’ve done both - for PLG as well as sales-led motions. For PLG, I was able to figure out positioning and plays to target customers in-product for plan upgrades and upsell motions based on their usage patterns, and verifying that with regional sales folks to understand what works.\n\nFor sales-led motions, I was able to identify if an increase in traffic = ICP vs. non-ICP visits via SwanAI  led to more demo form fills or not. SwanAI helped us profile, enrich and classify web visitors and ping us on slack with an outreach drafted. It was a good way for us to validate our SEO / AEO bets. Additionally, having access to data helped us reverse-engineer our ideal deal / account and customer profile based on past-won deals data to better target them for future pipeline.\n\n\n\nDirection still needs judgment\n\nThe premise is simple: being insight-led vs. data-driven is a lot better because it helps you drive cross-functional influence and evaluate tradeoffs better. Everyone talks about velocity, few understand it. Velocity is speed + direction. Speed is mostly solved for by AI. Direction is still very judgment and context-heavy, as most insights still need to pass the human smell-test.\n\nBeing insight-led also helps you ruthlessly prioritize closest to revenue tasks. I’ve said this before too  organizations are now becoming flatter by the day, and so every project invariably becomes cross-functional because skill gap has collapsed due to AI.\n\nThe playing field is quite leveled."
    },
    
    {
      "type": "post",
      "title": "Too much parenting, too little adulting",
      "slug": "too-much-parenting-too-little-adulting",
      "url": "https://www.varunchoraria.com/too-much-parenting-too-little-adulting/",
      "date": "2026-06-20",
      "tags": ["management","leadership"],
      "excerpt": "In the age of agentic AI, something I’ve first-hand witnessed as a people manager is that the default mental models are unchanged.",
      "content": "In the age of agentic AI, something I’ve first-hand witnessed as a people manager is that the default mental models are unchanged.\n\nWhat I mean is, a manager winds up oscillating between people issues and project management. Over time, this leads to significant burnout, and treating your team on the basis of pareto principles: top 20% don’t get laid off, others are victims of “paper trails.”\n\nPolitics invariably will happen if > 2 humans are involved in anything. An easy way to usually solve this problem is to use frameworks like RACI, DACI etc. and assign clear ownership. While it does work, you’re not fundamentally letting your reports function as adults over time. A manager then gets terribly burnt out chasing follow-ups and then finger-pointing on the basis of who lacks agency and who does not.\n\nOne fundamental truth that I’ve learnt in the last ~6 months is that you cannot expect everyone on your team to have an equal level of skin in the game. Not everyone will be engaged at the same level as you might be.\n\nSo, even in the age of agentic AI, you become the usual middle manager who is fighting to justify business value.\n\n\n\nBehaviors don’t change\n\nIn a diverse team, everyone comes from diverse backgrounds, various societal hierarchies, value systems and traumas. I’ve been in the business long enough to realize that no one is exempt from previous job PTSD and/or burnt by leadership.\n\nThis is especially visible when you inherit a team. The “new sheriff in town” aka, “my way or the highway” or enforcing SOP rarely works sustainably, since it is directly tied to how motivated the team is. And the moment business pressure increases, motivation tanks and everything breaks.\n\nSo then the convenient thing is to point at “behavior issues” of people and PIP them.\n\nThe net effect from a business equation perspective is that you’re probably at a worse spot than before you started.\n\n\nObserve and Absorb\n\nSomething I’ve realized is that most leaders aren’t fools (although some terribly are) and genuinely care about their team. So as a manager:\n\n\n  \n    Solve for intent first. In my experience, intent comes down to buckets like: documentation, process, budgets, cadence meetings, and retrospectives. Transparency needs to be the status quo; no information sits in silos. Lead through vulnerability.\n  \n  \n    Process: This is a tricky one. The current state might be such that file storage is one place, tribal knowledge exists at varying levels depending on team tenure, documentation might be an afterthought, planning happens in a different tool, reporting is in silos and heavily correlation-based, hygiene is a fever dream across the board. The common denominator of this is that process workflows typically sit outside of the daily tools and habits of the team. As a manager, using agentic AI, it is imperative to ensure whatever process you set sits within workflow. An example of this is that all campaign planning, monthly cadences, OOOs, etc. sat on a team google calendar with meta data enriched in the description. Leadership and cross-functional teams had access to the marketing team calendar, and were advised to look at it. Leadership did not have to shoulder-tap for basic information or navigate a labyrinth of google drive files or email threads.\n  \n  \n    Alignment: This is extremely inter-personal. Interview your team and get to know them as human beings. Your prime job is to enable them, and teach them to enable their stakeholders. Servant leadership needs to percolate in their daily habits as well. Why? Because trust is what gets work done. You do not need to build trust across the organization immediately. You need to build it in smaller pockets first, and work your way upward:\n\n    \n      \n        Instruct the team to write well-written emails, anticipating all objections first. A simple show-and-tell on leveraging AI to do this works wonders. You then become less of a manager-report, but more of a player-coach model which scales automatically. Habits like sharing minutes of meetings, keeping things in writing / async, going into meetings with a clear agenda written in the calendar invite itself, and flagging blockers early drives alignment in a healthy way.\n      \n      \n        Teach your team to say No. I do this by just clarifying the closest-to-revenue tasks. One way to share this mental model is that while everything can seem important - some are one-way doors and some are two-way doors. Or, some tasks are glass balls (which can break) and some are rubber balls (which can bounce back). All tasks can be done, but queueing can be prioritized to ship needle moving items every week - which can be attributed for. One practice I enforce in the world of MCPs is that now everyone has information-parity. So the expectation is to leverage that, create an air-tight business case and email that over for a quick spitballing exercise. I do not see much of an operating difference between product and marketing teams - both are in service of sales from day zero.\n      \n      \n        Share feedback on the problem-statement. More often than not, feedback - even the trivial ones, become a mud-slinging “he said, she said” fight which hurts credibility. Ultimately, no one can operate in silos so feedback needs to be entirely emotionless and tied to impact, justified by an SLA. You need to structure your message in that way, so you don’t leave room for interpretation.\n      \n      \n        Encourage self-promotion. Promotion in any org is always biased on showmanship and pompousness. Unfortunately, you cannot change the world but you can bring a certain level of meritocracy to it. Everyone’s work is important to achieving business goals, so why is each team member rewarded unfairly? I’m still figuring this out for myself, and something I’ve arrived at so far is you need to change the mechanics of reward. Like Mckinsey’s airport test, you want to encourage rewards for being a good person first. Skill / will gap can be solved for with coaching. But, for long-term growth and ensuring team environment stays positive - you want to reward things like EQ, empathy, ability to be coached, and how calm they can stay when shit hits the fan.\n      \n    \n  \n\n\n\n\nThe bottomline is that as a manager, you have to build an environment where people behave like adults as a function of time. That comes when you micromanage less, and instead resort to engineering acceptable behaviors that build the team culture."
    },
    
    {
      "type": "post",
      "title": "Killed By Google",
      "slug": "killed-by-google",
      "url": "https://www.varunchoraria.com/killed-by-google/",
      "date": "2026-06-16",
      "tags": ["career","strategy","ai"],
      "excerpt": "I first came across killed by google and was honestly quite blown. I suppose the startup myth is real: your instinct is right 95% of the time, but your ideas are wrong probably 80% of the time.",
      "content": "I first came across killed by google and was honestly quite blown. I suppose the startup myth is real: your instinct is right 95% of the time, but your ideas are wrong probably 80% of the time.\n\nYou only need one shot to make a real difference.\n\nAs I started to think deeply about it - both at a personal and company level - a few questions came to mind:\n\n\n  Is it even feasible if you don’t know where your next meal is coming from?\n  Is it feasible to tokenmaxx your way to vibe coding a million little ideas in a company struggling to meet ARR goals?\n  Is it even worth the time and effort to experiment?\n\n\nStripe also apparently had # crazyideas back in the day where anyone in the company regardless of seniority could build an idea the company should be prioritizing and talk about it publicly. I’ve seen versions of these morphed as hackathons in the handful of companies I’ve worked at.\n\nIt’s a good move, but in an age where everything comes at a cost, is there light at the end of the tunnel?\n\n\nUtter Chaos\n\nRamp is fundraising to cover token costs. Freshworks, where I worked at previously, let go 11% of their workforce in 2026 because most of their code is now written by AI.\n\nSo how does one have fun and build something meaningul in an age where everything is as chaotic as it gets? Let’s face it - motivation has a ridiculously low mileage, and frontier models are costly especially if you’re outside of the US and can be gatekept easily like Fable 5.\n\nMore importantly, you might not even have access to a decent computer to work on.\n\nWhat’s the way out?\n\nWell, in my experience, the fundamentals don’t change and in an increasingly chaotic world (it was always akin to brownian motion, in my opinion) - the fundamentals become even more important.\n\n\n\nThe fundamentals: Relationships + Side Projects > Resumes\n\nI have zero idea why resumes were created in the first place, but for the sake of this conversation let’s assume an old IT uncle wanted to add extra hoops to employment.\n\nWhich, ironically, was then pre-screened by biased ATS software and was never read by unc.\n\nThe world always disproportionally rewarded curiosity - it’s just bubbling up to the surface now. Regardless of the software segment - the top players are consolidating (Fin being acquired by SFDC in 2026) to protect margins while anyone in the low or middle part of the spectrum are chasing capital efficient growth.\n\nNot because their CFO said it was the smart thing to do - but burning money post ZIRP became that much costlier to do so. ‘Growth at all costs’ mindset soon morphed into a fight-for-survival, powered by AI at the cost of human labor.\n\nAt the end of the day, career fulfillment comes from working on the work you love, and with the people you admire. After a point compensation can only motivate you so much.\n\nIn an earlier post, I gave my version of how unfairly high the bar is now. Earlier you had 5,000 other people applying for the same position. Now you’re competing against AI too. To be fair, the world always worked and recognized the same things: good judgment + good craftsmanship = rubber hits the road. Ideas are a dime-a-dozen anyway (quite literally since many indie hackers are monetizing it)\n\nSo, you can do a couple of things:\n\n\n  \n    If you have access to AI at work - regardless of what type (ADE like cursor, LLMs like Claude or codex) - do the work once, automate it the next time, and then document it on your blog or newsletter or linkedIn. That’s how you build a rubber-meets-the-road portfolio without throwing in a personal dime. You’re also simultaneously building a brag doc for workplace recognition.\n  \n  \n    If you don’t have access, opencode is freely available, along with models you can download off open router along with skills on github you can build on top of for side projects.\n  \n  \n    Once you build and explore a few things, reach out and comment on people you admire on X or LinkedIn. Exchange notes, build relationships. You’ll waste less time in the process and actually get direct access to someone’s eyes and ears. You’re engineering serendipity this way and opening up collaboration.\n  \n\n\n\n\nIt’s not hard to do meaningful work, but it’s not easy either."
    },
    
    {
      "type": "post",
      "title": "Who owns what?",
      "slug": "who-owns-what",
      "url": "https://www.varunchoraria.com/who-owns-what/",
      "date": "2026-05-17",
      "tags": ["ai","management","strategy","leadership"],
      "excerpt": "Everyone these days is asking the same question in the age of agentic AI: what should an AI do vs. what should a human do?",
      "content": "Everyone these days is asking the same question in the age of agentic AI: what should an AI do vs. what should a human do?\n\nSome background first.\n\nI remember last year (2025), in a previous org, I championed the use of customGPTs to break down knowledge and skill gaps within the product marketing team. The problem statement by leadership was simple: writing is core to product marketing, but not everyone is good at it.\n\nThe level 1 of solving that problem was to fix the technical writing side of things: it’s hard to memorize use-cases, JTBDs of every new feature and enhancement released. Knowledge base / feature documentation is not always hygienic, and even if it is — there’s a lag between when the latest and greatest is actually published.\n\nProduct marketing or even marketing is a frontline engine, and therefore needs to be the first to create the buzz in the market. Not having access to the latest technical docs can directly hinder the freshness of any collateral produced.\n\nI owned training the customGPT via .md files because it could ingest a lot more data. Remember, this is a time when even the team plan on ChatGPT had limits to how much you could load. Older models like 5.3 and 5.4 could only cache so much and context windows would be tiny.\n\nTechnical problem — mostly solved.\n\nNow, the creative writing problem. When you’re in a larger public-listed product company, typically your writing needs to go through legal before it sees light of day. Besides being messaging-aligned and keyword-infused, writing had to be legal-approved as well.\n\nThe answer was guardrails. I instructed the customGPT to read certain .md files first which had the best-practice guardrails intact. The GPT would parse those files first, ask clarifying questions to an open-ended prompt like “Write me a landing page” and work with the user to build the content. It would then end the output with checks on guardrails.\n\nIn between, the user would have the autonomy to write in a tone they preferred.\n\nAcross a 40-member product marketing org spread across 2 business units, the quality and velocity both improved in under a month.\n\n\n\n\n\nThe evergreen question\n\nNow in the agentic AI world, where everyone is using MCPs, skills, agents and subagents — things have become more advanced as AI percolates into deeper and more complex workflows, in addition to job descriptions themselves being re-written.\n\nThe question still hasn’t changed: Who owns what?\n\nThe single mental model I’ve shared with my team at GTM Buddy was this: do the work once, automate it the next time. Instead of starting at absolute 0% next time, try pushing the start line to 20-30%. Over time, 100% will look a lot better in terms of both quality and token cost.\n\nThe older mental model from pre-AI about a job was that it’s a static list of tasks, and hiring an agency or a full-time person is a way of outsourcing it. AI will replace the task, therefore the job. Much of the world is still operating on that mental model, whether they like to admit it or not. The clearest signal is them cutting down on headcount (easiest lever to pull) and squeezing more out of the remaining few. Up until now, the few at least had access to nearly limitless tokens, but in H2 of 2026 and beyond — the “get more done with less” will also seep into your AI usage.\n\nThe better mental model to go after this understanding what the JTBD is, not just the job:\n\n\n  Jobs = tasks + judgement + context + ownership + metrics\n\n\nThis changes the hiring equation quite a bit, because EQ becomes a signal you look out for in candidates, and give lesser weightage to technical abilities. In the pre-AI world, most people who climbed the corporate ladder where inherently great at the people equation: change management, relationship building, taking solid judgement calls, and stepping up when things break.\n\n\n\nThe evergreen answer\n\nAs anyone in the management side of the business, you need to stop asking yourself this question:\n\n\n  Can AI do the job?\n\n\nAnd continually ask yourself:\n\n\n  Does a task require context that can exist only between 2 humans? Eg. managing an executive relationship between a seller and a prospect.\n\n\nIf the answer is:\n\n\n  Yes → Human or human + AI owns it\n  No → AI owns it\n\n\nEvery quarter, as a management person, you need to sit down with each team member and work on updating a document around each person’s swimlane.\n\n\n  For each swimlane, list down the entire process. Specify each task in the process, and classify it as owned by AI / Human / Both. Alongside it, also write down what’s happening in each task, and what to do / what not to do.\n  This needs to be visited every Q, as a larger human + AI retrospective. This is what the new coaching would look like in the agentic AI world, between a manager and a direct report. It prevents skill sprawl, using AI as a hammer to every nail, and encouraging smarter AI usage.\n  For the entire team, the manager can draw out and publish a boundary map: list the entire piece of the business a team impacts, swimlanes under each, and task distribution between AI and Humans. If you’re able to tie it to growth metrics — that’s a validation that you’re truly using AI to drive real business outcomes, not just becoming AI-native for the sake of it."
    },
    
    {
      "type": "post",
      "title": "Team values I value",
      "slug": "team-values-i-value",
      "url": "https://www.varunchoraria.com/team-values-i-value/",
      "date": "2026-04-25",
      "tags": ["management","leadership","strategy"],
      "excerpt": "We work with a bunch of vendors as a company of 44 people. As a manager, I have 2 choices:",
      "content": "We work with a bunch of vendors as a company of 44 people. As a manager, I have 2 choices:\n\n\n  I can coach my team so they can manage vendors and get the desired output.\n  I can manage everyone myself, but at the cost of foresight.\n\n\nThe second choice keeps control with me and taxes foresight. The first distributes judgment. In a lean, agentic organization, that distinction matters: the manager’s job is not to become the bottleneck with better dashboards. The job is to turn people into operators who can manage vendors, AI agents, and stakeholders without waiting for permission.\n\nManagement is something I’ve had to figure out. Books are great for anecdotes but seldom translate to the constraints of the real world. Especially your unique constraints.\n\nThe problems each team faces aren’t entirely unique. They are usually variations of the same patterns. But the solution and the sequence of execution are unique in practice.\n\nI went with option 1. I decided to give my team the playbook of how I manage, and coached them on how they can onboard, ramp, and manage vendors.\n\nIn the age of AI, I doubt there will be more than 2-3 layers to an organization. The upcoming era of grooming the next generation of leaders will be centered on managing AI agents, vendors, and stakeholders, not direct reports.\n\nSo here is everything I value as a manager, from a cultural perspective. This is not a motivational values doc. It is an operating standard for how I expect people to behave when the map is blank.\n\n\n\n\n\nWhy is the map blank?\n\nI ran a podcast in 2020-21 during the COVID years and first-hand witnessed how HR leaders navigated a world without a playbook. Transitioning an entire organization to remote work overnight is no easy feat.\n\nRunning a remote team is extremely high-trust. I was pretty amazed at how the HR function was able to bring about this transformation with minimal business disruption.\n\nThat’s when I realized that it’s not just the rules, but the battlefield itself that changes. Constantly.\n\nUp until 2025, the problem management faced was having a spectrum of workers on the “shop floor”: Gen X, millennials, and Gen Z.\n\nAs agents have become the norm in 2026, you have agents and AI tools in the mix too.\n\nSo, how do you define team culture and strategy in the agentic era?\n\nThat’s why I feel that the map is perennially blank. The goalposts change, the metrics change, the work changes, and the composition of the team changes. Earlier, I used to measure individual performance by productivity metrics. As a manager now, I’m looking at average revenue per employee.\n\nEveryone’s OKR is revenue-focused. Causally. Not through loose correlation. If you cannot draw a line from your work to revenue, the work is probably theater.\n\n\n\nWhat culture really means to me\n\nCulture is the DNA of how you operate as a team and as a company. It’s the playbook, the modus operandi. Benefits and perks aren’t culture to me. It’s how the team consciously chooses to operate without supervision, and how information flows both within the team and outside of it.\n\nIt’s a code of conduct, but in the context of what the individual and the team collectively stand for.\n\n\n\nMy non-negotiable values\n\nOwnership\n\nTo me, this simply means 2 things: following up and following through. The way I differentiate between the 2 is:\n\n\n  \n    Following up = how proactively one is sharing updates, surfacing blockers, and thinking from a systems and first-principles perspective for every task. I do not like order-takers. Did you close the loop without being chased?\n  \n  \n    Following through = can you see a task through end-to-end and holistically? The thing with ICs is that sometimes one can be too focused on their swimlane and forget to zoom out and see how it works in conjunction with the larger goal and the rest of the business.\n  \n\n\nAccountability\n\nAccountability is not a task list and blind update-sharing. Success is a collective responsibility. Here’s what it means in my books:\n\n\n  \n    What is your say/do ratio? I wrote about why consistency compounds as well. This is disproportionately important in my team.\n  \n  \n    How proactively can you manage stakeholders? In my opinion, managing stakeholders is not rocket science. It’s a simple recipe of an email trail, removing blockers, tactfully enabling people, and knowing when and what to escalate. I always encourage my team to work with a mindset of making your stakeholder’s job easier from day zero, without expectations. Send them templates, content, and anything else that’s needed without even asking. Proactively and periodically ask them for blockers. No one ever opens up about it without a nudge.\n  \n  \n    How do you balance short-term vs. long-term fixes? Projects are inherently grey areas. In the interest of shipping things, short-term fixes are great. But can you figure out a way to implement a long-term fix by yourself, so you don’t encounter the same problem twice?\n  \n  \n    Do you know how to say no, both upward and to your stakeholders, without a personal agenda?\n  \n\n\nAgency and Urgency\n\nI would’ve called this “work ethic” in 2025, but in the agentic era, it’s a rebrand.\n\n\n  \n    Why not today? Decisions are either one-way doors or two-way doors. One-way door decisions need to be made carefully, deliberately, and in compliance. Two-way door decisions need bias towards action. ICs that own both the outcome and the decision become true thought partners. The best feedback loops in the world are action-driven, not document- and approval-driven. Validate and invalidate hypotheses, and let data and market sentiment guide you. Not a BS approval matrix.\n  \n  \n    How ambitious are you? It’s one thing to scrape a prospect list, another to tier and score it, identify buying committees, categorize them into what funnel they need to go into, and then build a personalized, automated engine that runs on its own. The human-in-the-loop aspect needs to be minimal, and on a cadence.\n  \n  \n    Do you know your customers’ pains and problems? I wrote about the difference between them and it’s mission-critical for success. It gets you respect super quick across functions too. Knowing your customer well, and how their needs are evolving, is foundational to understanding why you choose to optimize for a certain north star metric. If prospects and customers are finding it hard to buy from you easily, there’s no point in losing sleep over site impressions.\n  \n  \n    Are you tied at the hip? 11/10 teams work in silos. Even in the same team, one does not know what the other is up to. Over my career, I’ve seen this quickly evaporate all remaining empathy. Secondly, I’m a huge fan of theory of constraints, and it is the single mental model I use to solve any personal or business problem. As a manager, I intentionally ensure cross-pollination by getting individuals with complementary skill sets to team up and own an outcome. It’s always a driver-copilot model in every team, but no more than 2-3 folks per project. Beyond that, mimetic thinking and herd mentality set in.\n  \n\n\nCraftsmanship\n\nBe enviably good at your job. It’s a professional obligation to stay at the frontier of what’s happening in your field. In everything you do, the operating mindset needs to be:\n\nCustomer > Company > Team > Self\n\nDocument, and document everything\n\nContext is everything. Documentation should be a default mindset. How well you write, format, and organize your documents will tell me more about how far you’ll go in life than the skills you’re probably arrogant about.\n\nEarn your seat, every day\n\nI hire slow, but fire fast. No passengers. A team is like a high-performing sports team. If you want to get drafted for the next season or want to retain your position, then you have to earn your seat every single day.\n\nKindness is default. Diplomacy is defensive.\n\nOne thing I’ve always told my team is that you have to operate with a blind faith that your manager and leaders have the best intentions for you at heart. Even when they’re giving feedback that is hard to swallow.\n\nYou’re not competing with your peer. In the battlefield, you’re responsible for each other’s success.\n\nNothing erases your entire credibility faster than being defensive.\n\nAnd more importantly, have a sense of humor. I love taking jokes on myself, and it’s a non-negotiable to have fun along the way. Nothing is worth it otherwise.\n\nSay it, or forever keep your silence\n\nSay it in the room where it matters or in Slack channels. But sharing your real opinion after the fact really doesn’t help anyone."
    },
    
    {
      "type": "post",
      "title": "I've been trying to get lucky for a decade. Here's what I learnt.",
      "slug": "ive-been-trying-to-get-lucky-for-a-decade",
      "url": "https://www.varunchoraria.com/ive-been-trying-to-get-lucky-for-a-decade/",
      "date": "2026-04-19",
      "tags": ["career","strategy","leadership"],
      "excerpt": "I think one of the things I have a perpetual obsession with is career optimization. It feels a lot like poker. There is no magic playbook, or a clean meritocratic path to it, at least not anymore.",
      "content": "I think one of the things I have a perpetual obsession with is career optimization. It feels a lot like poker. There is no magic playbook, or a clean meritocratic path to it, at least not anymore.\n\nLuck is taboo, though, and it takes a certain level of maturity to come to terms with it, because it is the variable that never really leaves the equation of career and life.\n\nSo instead of learning the next AI tool or chasing the next gig for a 5000% hike, why not focus on improving your odds?\n\n\n\n\n\nDo, tell, or both?\n\nI’m a fan of mental models. In a highly unpredictable world, even an NPC needs waypoints and checkpoints to find the most efficient path to the destination.\n\nOne mental model I’ve come to really respect ever since I became a manager is doing and telling at a 50:50 split. I’ve seldom seen only talkers or only doers succeed for very long, but a combination of both almost always does.\n\nSo if you were to plot “Do” and “Tell” on an X-Y axis, the slope of it is luck: the right amount of doing combined with the right amount of telling. That’s how you create any sort of luck, whether it is getting hired from shitposting on X or LinkedIn, bypassing the snake pit of HRs to talk to hiring managers directly about a job, or getting paying users. The list goes on.\n\nIn my world, doing and telling has gotten me some pretty interesting roles to work in. The latest one involves managing a team and helping build a product category. So if I look back at the career bets I made:\n\n\n  job #1: Series C startup -> founding product marketing team member -> scaled it across India and the US. First touch with enablement, no-code tools, design, and product.\n  job #2: IPO company -> started the PLG motion -> scaled it to half a million dollars -> learnt the nuances of scale and company politics.\n  job #3: Series A startup -> helping build category -> managing a team -> increasing ACV, win rates, pipeline volume and quality, and founder brand.\n\n\nThe point of telling you this is simple: I’ve never posted much online, but I do a lot of doing and telling internally, and that has helped me get into whispered roles, the kind that don’t really have job postings. So you do not necessarily have to be a social butterfly to increase the slope, or the surface area, of luck in your life.\n\n\n\nWhat’s after luck?\n\nDestiny.\n\nLuck is inherently probabilistic in nature, but let’s face it: gambling takes a lot of energy, and over time your risk appetite reduces as you grow older. There is a short window to make anything meaningful happen before your destiny becomes clear to yourself and to the market.\n\nSo how do you become what you are destined for?\n\n\n  \n    To begin with, don’t necessarily job hop for hikes. Or if you’re running a business, don’t make pivots a habit. Asymmetrical bets have asymmetrical rewards, and the common denominator is persistence. Persistence to your craft, to yourself, and then to the company you work for, aka tenure. Become unbelievably good at what you do, create side projects, and make sure you have more than 10 all-weather industry friends. It’s easier to make friends than people think. Ask good questions and make keen observations. That’s one of the fastest ways to earn respect and credibility with senior folks.\n  \n  \n    There is no substitute for action. There is a lot of grey area here, but let me condense it: action is both art and science.\n\n    \n      Art, because you have to develop taste, bring polish to your deliverables, avoid bringing one option and instead always present alternatives with trade-offs, follow through without being chased, and spot purple cows, which is just another way of saying spot what others miss.\n      Science, because the boring questions still decide a lot. Did you do what you said you would do? How do you correct yourself when you’re wrong? Why go with tool X instead of Y? How will you automate this and do it better next time?\n    \n  \n\n\nYour work ethic is the single biggest lever you can pull that determines your destiny. Work for the love of work, and the rewards will come in due time.\n\n\n\nWork-life balance, not having money, not having access to the right people, and other BS\n\nHere’s what I tell, train, and coach my own team on:\n\nEveryone has constraints. But like players on a sports team, you have to do a lot of painful soul-searching to find your superpower, or your edge. I often ask this in interviews too: “What’s your superpower?” It could be anything, making the best pancakes, knowing exactly how a rotor works, literally anything.\n\nIt’s a real signal on obsession, depth, and something a person spends serious self-directed time on. Those are transferable skills.\n\nMy team, a designer and a GTM Engineer, are getting quite skilled at adjacent skills because of AI. I wrote about that here. I’m quite big on them building their personal brands and doing a lot of do and tell on the internet. Here’s how I’m coaching them:\n\n\n  Document every single project you shipped through the lens of the business problem you solved. Context is shared transparently with them, and I weigh managerial and project decisions openly in team calls. Nothing meaningful sits on the other side of the boundary. Everyone is an equitable stakeholder.\n  If documenting is a pain because time on weekdays is scarce, literally open drafts in Gmail and write bullet points. Or maintain a brag doc that you update weekly.\n  Take real screenshots of what you built, the constraints you encountered using your toolsets across projects, and introspect hard on what worked and what didn’t. Identify patterns over time and take strategic calls every quarter: retire a tool, redefine a process, or establish a different way of communicating with stakeholders.\n  The only true way to grow is to compound. All other growth is incremental and linear. Flatlined careers are dead careers. So compounding is easy, at least in theory: self-review and introspection.\n\n\nYou don’t need access to specific people, an executive coach, the right computer, or the right tools. Work-life balance is a great fever dream but with no meaningful implications.\n\nYou don’t even need the right curriculum. Lord knows I’ve spent a lot of time curating one for myself over the past decade. But you’re no professor. You’re a doer. It makes more sense to tactfully solve the problem in front of you and learn from it."
    },
    
    {
      "type": "post",
      "title": "Imposter syndrome as a first-time manager",
      "slug": "imposter-syndrome-as-a-first-time-manager",
      "url": "https://www.varunchoraria.com/imposter-syndrome-as-a-first-time-manager/",
      "date": "2026-04-12",
      "tags": ["management","leadership","career","ai"],
      "excerpt": "It’s been about a quarter since I became a manager. I’m wondering what value I’m actually adding to the organization, especially in the agentic era where skills, information, and knowledge are more or less commoditized. Here are my candid observations.",
      "content": "It’s been about a quarter since I became a manager. I’m wondering what value I’m actually adding to the organization, especially in the agentic era where skills, information, and knowledge are more or less commoditized. Here are my candid observations.\n\n\n\nPre-AI\n\nI remember in 2025, I was part of a 5,000-person organization, and I was an IC3 — 3 levels from the junior-most rung. Above me, we had 6–7 layers until the CEO. Add the complexity of multiple BUs, conflicting priorities, random OOOs from stakeholders, 25,000 meetings per week, and 7 different “leaders” and their friends.\n\nIt would take about 2 weeks to stitch context, rally the right stakeholders, convince them that project X will get them promoted, figure out the escalation matrix, SLAs, project timelines, dependencies, and so on. And if your luck is genuinely bad, the project’s scope will either get changed at the end of week 3, or someone will sidestep you and take the lead mid-way, or you’ll be questioned on ARR impact despite the project being a “P0” at the time of launch.\n\nAll of this to change the color of a button.\n\nI figured out a couple of things amid all this busyness:\n\n\n  Hierarchy is not about authority at all. Until 2025, it used to be about information routing — being a “router” like the WiFi router. You’re just aggregating, synthesizing, and relaying stuff, mostly because the organization is too big for anyone to see everything. Then layer in project updates, QBRs, stakeholder reviews, and leadership reviews, because everyone eventually forgets what they were working on or why, due to the complexity.\n  Every manager mostly made money because routing information to the right people at the right time got that much more expensive as you ascended the corporate ladder.\n  Every BU was also bloated because every individual brought in adjacent skill sets. It’s not that the people weren’t skilled, but moving information from design → prototype → engineering → QA → alpha launch → beta launch → GA required about 50–100 people to be in a tight handshake.\n\n\nSure, design → GA launch of a feature can happen in under 2 months. It was never about speed. It’s the linkages between stages that were the real cost — translating what the designer intended for various JTBDs into the PM’s worldview of business and edge cases, into engineering constraints, and so on. That’s where the most energy per unit time was lost.\n\n\n\nPost-AI\n\nLean teams are going to be the status quo sooner rather than later. I lead a team of 3 to help generate $XM pipe per quarter. Prior to 2026, this would be the job of 6–7 people and would cost a lot in terms of ad spend. Time to first meeting and stage conversions would also be an afterthought, because all your focus is going into generating demand or plugging holes within the demand engine in a reactive manner.\n\nAI and agents have fundamentally democratized skills. Everything is transferable now, and information and context are no longer the bottleneck. As a result, speed of shipping quadruples.\n\nThe designer on my team can design, do keyword/AEO/GEO/SEO-optimized copywriting, and take things live in under 2 days — including manager and leadership reviews.\n\nThe GTM engineer on my team can quickly query the design and campaign plan and get instant context. In a couple of prompts, he’s able to build the data pipeline, routing logic, customize CRM fields, and scrape and enrich ICP information for sequences and ad targeting. All of this in under 2 days.\n\nBoth work with virtually zero meetings between them.\n\n\n\nMy imposter syndrome and how I fixed it\n\nOne of my mentors once said that the 3 hardest career transitions are:\n\n\n  IC → Manager\n  Manager → Director / VP\n  VP → C-suite\n\n\nAs a manager, I initially started as the router. I was relaying, aggregating, and contextualizing information for my team — passing the torch between my designer, GTM engineer, and performance marketer to keep the project moving.\n\nTowards the tail end of Q1, I asked myself a hard question:\n\n\n  “If everyone else is doing the work, and I’m just doing air traffic control, what am I even being paid for?”\n\n\nThat’s when lightning struck.\n\nI report to the CMO, and she said one thing:\n\n\n  “Everyone should be a full-stack marketer.”\n\n\nThat sounds like a great anecdote, but it’s a massive mandate in the age of AI. Here’s what I’ve realized about what a manager’s source of value actually is in the agentic world:\n\n1. An agentic manager blurs the lines between roles.\n\nWithin the marketing and larger GTM organization, the job is to remove the friction in handoffs, alignment, and context-sharing — not to be the one who orchestrates it manually, but to build the conditions where it happens organically.\n\n2. An agentic manager must end silos.\n\nIf there’s one thing I hate, it’s siloed work. Nothing great has or will ever come out of it.\n\nIn the past, stakeholders in a project worked sequentially:\n\n\n  copy → design → landing page optimization → scraping / building lists → paid and organic media → reporting → campaign optimization\n\n\nIt seldom started with problem statements, baselines, or deliberate bets. Debates between teammates were purely operational, not strategic. No one learns anything new. Campaigns never compound, and you reset the clock every quarter instead of building on the previous one.\n\nAs a manager, what I need to enable is parallel work on the problem statement, aligned to the individual capacities of any given campaign.\n\nEvery campaign is unique, even if you’re hitting the same target account list. Your method of distribution is different, your offer is different, your campaign goals are different. My success as a manager largely depends on my team’s outputs.\n\nIt makes zero sense to keep hitting operational roadblocks due to a lack of transferable skills or shared context across campaigns. There’s no marketing leverage being built. After a decade in marketing, I’ve realized that marketing is a lot of grunt work. AI reduces that — but it doesn’t eliminate it.\n\nThe real leverage comes from aligning capacities to skill sets. A live example: we were trying to improve form conversion rates on our website — the single biggest driver of leads. I got my designer and GTM engineer to spend 70% of their time building an airtight business case document (no meetings) covering:\n\n\n  Audit of existing UX issues, benchmarking best-in-class, and an RCA on why companies across revenue bands went a certain direction — and whether it’s relevant to us\n  All the paths that ICP vs. non-ICP visitors take, along with source, sessions, and performance data for those pages, so we know where to double down to reduce time from homepage visit → form fill\n  Form page metrics: sessions, engagement, bounce, etc.\n  Form completion → time to first meeting\n  Qualification, scoring, and routing for all variations of visitors\n  Meeting prep for the SDR team\n\n\nThe humans spent the bulk of their time getting clarity on the problem statement and thinking holistically. All of this serves as the base context before anyone opens their LLM window blindly. In parallel, the team picks up those problem statements and works with AI, MCPs, and skills to build a deep understanding of the 6 steps above. Once everyone has the same level of clarity, execution happens at high velocity. The designer isn’t dependent on the performance marketer for data. The GTM engineer’s POV never conflicts with design direction.\n\n3. An agentic manager has to kill the “big launch identity.”\n\nI’m a proponent of continuous improvement. Big releases do happen, but we don’t live and die by them. We live and die by meaningful daily ships that move the needle month over month and quarter over quarter. Every week, the cadence is to tighten the problem statement, self-identify things to improve across projects, and ship with a common context foundation, powered by agents and AI.\n\n\n\nHow we actually work\n\nSkills repository\n\nThe marketing team works out of GitHub. We forked skills from marketing-skills.com and skills.sh, reverse-engineered outputs from other teams, contextualized them to our organization’s needs and constraints, and built on top of them.\n\nYou don’t necessarily have to start from scratch every time.\n\n\n\n\n\n\n\n\n\nI constantly maintain this repository and ship skill and workflow updates every week. I also do a virtual lunch-and-learn with the team on how I’m using it, and how I would if I were in their seat.\n\nChange management must be led from the front.\n\nWorkflows\n\nWe use SwanAI for automating GTM workflows from Slack — deanonymizing website visitors, crafting personalized hyper-contextual outreach messages (automating SDR work), account research, deal and pipeline reviews, sending data to CRM, and much more. We do a lot of “Vibe GTM” and prompt our way through ensuring GTM runs smooth.\n\n\n\n\n\n\n\nWrapping it up\n\nAs an “agentic manager” — and this holds true for any manager, anywhere — my roles and responsibilities are to:\n\n\n  Coach and elevate performance. The leading indicator for me is that ICs work on increasingly complex problem statements over time.\n  Work once, automate it next time. The team has a clear goal: never do the same work twice. The team is incentivized to automate their work at the end of every project. If you started a task at 0% last time, you have to start it at 40% the next time.\n  Only work on tasks closest to revenue impact. Say no a lot more. A key goal I’m constantly working on with the team is sharpening their judgment so they know what requests to pick up and what to push back on.\n  Average revenue per employee is the one metric that matters. Everything else follows. If my designer is working on redoing the website, they’re required to make design decisions reverse-engineered from what will move the revenue needle. Aesthetics and vibe need to be the consequence, not the driver.\n  Reframe the reward. The dopamine hit needs to shift from promotions and titles to how the market responds and whether you’re hitting your revenue goals.\n  Documentation is everything. Every month, I train my team on skills. For Q2, it’s stakeholder management and building a business case. The ask is to write a detailed, data-backed document and email it to me. I find ways to shoot it down. They push back with stronger arguments. We spar on email — this improves multiple skills, especially managing upward."
    },
    
    {
      "type": "post",
      "title": "So you want to help your team succeed",
      "slug": "so-you-want-to-help-your-team-succeed",
      "url": "https://www.varunchoraria.com/so-you-want-to-help-your-team-succeed/",
      "date": "2026-02-07",
      "tags": ["management","leadership"],
      "excerpt": "It’s been about a month since I’ve been a first-time manager. Before I started, I asked my CEO what advice he had for me. Here’s what he said:",
      "content": "It’s been about a month since I’ve been a first-time manager. Before I started, I asked my CEO what advice he had for me. Here’s what he said:\n\n\n  Partner, don’t command\n  Don’t throw weight around because of the title. It’s alright if it happens unknowingly, but it’s crossing a line when done intentionally\n  Take up more work, because that’s how you build trust\n\n\nI understood these, because it was simple and straightforward. Coming from an IC background, this was sort of natural to me anyway because I didn’t know how to lead. But then, there came a point very recently when I realized the entire marketing team was tiny, 4 folks including me.\n\n\n\nThe problem with a tiny team\n\nThat’s a natural setup both in the age of AI as well as in a company at the Series A stage, but I was in a tug-of-war between 2 very difficult problem statements:\n\n\n  Marketing traditionally was not seen as a growth lever. We had an ambitious charter of both category creation as well as driving pipeline. Driving pipeline is a straightforward mandate, but not when you have to revisit your basics of ICP and lead qualification. Over and above that, we were also trying to cope with AI and automations, trying to do our job once and then automate it using skills probably the next time.\n  The second issue was about unlocking capacity of each team member. That meant better ownership, accountability, craftsmanship, and also becoming a strategic thinker. Being strategic at the outset for us meant one thing: what is our unique insight about both our swim lanes, as well as the market?\n\n\nWhen deadlines started becoming tighter and constraints more unforgiving, I was in a spot where I had to reach out to my manager, the CMO, to also wear an IC hat and step in. We needed all hands on deck because we were creating a whole new market category.\n\nTo my surprise, my manager said one thing: “Give me more work to do. Assign tasks like you would generally to your reports.”\n\nThis was unusual for me, since I always assumed you would only go to your higher-ups with solutions to problems, not work itself. Asks are different because they’re project and stakeholder dependencies. Here I was, coming up with a laundry list of items to work on.\n\nWhat servant leadership means\n\nThat’s when my manager introduced me to “servant leadership” as a concept. I loved it, and I’m in the process of internalizing it myself, but here’s what I’ve understood about it:\n\nA true leader’s goal is to serve those they lead. It’s kind of like taking the inverse of a problem you have and reverse engineering it. Because if you think about it, a leader’s real mandate is to drive influence, not direct and execute.\n\nAnd this makes logical sense to me, because if you think about it:\n\n\n  Maslow’s law dominates it all. If needs are met, belonging and growth in the context of career, meaningful contributions happen.\n  As an ex-IC, and I hard relate, system and work constraints bubble up from the bottom.\n  Motivation and incentives are tightly coupled. That’s when folks really go above and beyond, and I believe you need that in any setup.\n\n\nPutting it into practice\n\nAnd I’m unsure if I’m right in doing this, and time will tell, but here’s how I’m putting thought to action:\n\n\n  Whenever a task needs to be done or assigned, I always lead with one question: “what do you think about this approach?” It might sound simple, but my notion with that is to make everything we do a partnership. Decisions need to be democratic. Information needs to flow transparently, and bi-directionally, regardless of good or bad.\n  Building people first. I’m approaching this by guiding teammates to become more strategic. Take your task, automate it or make it into a dashboard, and then you’re responsible for sharing insights, correlations, and causations. All tied to core business metrics, not just what you handle.\n  Removing barriers. This is a tricky one. Barriers are often cultural, that manifest as technical. I openly conveyed that marketing was not conventionally seen as a growth lever, and it must. So any task we pick up needs to be closest and directly in service of revenue. And so when random work comes through from other teams, we clarify the need of it based on revenue impact. If focused effort becomes culture, then process becomes ammunition and a force multiplier, not a defensive shield.\n  Lastly, one thing another mentor of mine said is that being a leader is a very thankless job. You’re paid more to shield your team from shit, to bring clarity to the table, and genuinely invest in the growth of others. That means pointing the limelight to them. No one will ever thank you when BAU happens as intended, even though there’s a lot that goes into it."
    },
    
    {
      "type": "post",
      "title": "\"You're not strategic enough\", said Manager",
      "slug": "youre-not-strategic-enough-said-manager",
      "url": "https://www.varunchoraria.com/youre-not-strategic-enough-said-manager/",
      "date": "2025-10-26",
      "tags": ["leadership","management","strategy"],
      "excerpt": "Cracks",
      "content": "Cracks\n\nI believed for most of my career, if not all of it so far, that you had to be the best possible IC in order to become a manager or leader. Especially with the whole narrative around “the builder’s age” and “one-man marketing teams” or “office politics sucks”, the notion of being a lone wolf kept getting reinforced.\n\nI often thought one had to:\n\n\n  know their craft really well\n  be super-fast and error-free in all of their deliverables\n  have the entire company’s knowledge and context memorized\n\n\nand then magically, the promotion fairy would appear and place that XX% hike under your pillow and give you a kiss.\n\n\n\nCrevices\n\nSure, in the early days and years of your career, it is important to be a dependable workhorse. That’s what leaders and managers look for: can I count on you to get the job done, and well?\n\nAfter a couple years in the trenches, you’re bound to become a high-performer. Form follows function. But what’s happening in parallel is that there are others who have also become high-performing like you as well.\n\nThe mind then instructs one to take the “here’s how it’s done” approach. It’s essentially where you take the megaphone and keep yelling the best practices, hoping someone would hear and notice. Especially your leaders.\n\nBut this tunnel vision seldom scales. Today you might be a random employee. Tomorrow, you will have to band together people, create a vision, and deliver results.\n\nA complaint I’ve always had is that the transition from IC to leader is knowledge that is incredibly gatekept. Very few industry folks actually go the extra mile to train, coach, and mentor you. 99% of others are left to aimlessly fish around and find out. If they do find out, they’ll probably breed another generation of bad leaders.\n\nClosure\n\nWhat I’ve been able to figure out is that the journey from IC to manager or leader is essentially from “Here’s how it is done” to “Here’s how it is done so you can help me do more.”\n\nIt is a mindset shift from rapidly acquiring skills to enabling and uplifting others at scale. Uplifting others means:\n\n\n  sharing complete context and information in real time, not as an FYI\n  building a culture where you treat direct reports as thought partners, not as handymen to figure out operational feasibility\n  setting the bar high and coaching, which means:\n    \n      ride-alongs\n      showing what good looks like, and dissecting each part of it\n      actually mentoring people through that process by holding them accountable\n      not giving up after a few tries\n    \n  \n\n\nIn most cases, the norm is to tell direct reports what to do and get the job done as a team. It works great in a sweatshop where inputs and outputs are predictable enough.\n\nBut a huge downside of that is you’re secretly stealing away their ability to make decisions. The little micro judgment calls that let them close the loop on task ownership. It prevents them from being agile and innovative.\n\nAfter a point, managers become the ringmaster to a circus full of monkeys. That’s what most AI-powered crap-a-lytics platform companies are these days.\n\nAs an IC:\n\n\n  get into the habit of asking the best questions, not having the best answers. Zoom in, zoom out, see how it affects other stakeholders, and see if it’s aligned with both organizational charter and goals or not.\n  get into the habit of aggressively self-reporting in terms of impact. Move from “I did X, Y, Z this week” to “X led to $ impact, a % uplift from last quarter. To reach our target of double-digit %, A, B, C are blockers.”\n  get into the habit of building dashboards, or leveraging existing ones, or even Excel models. Do scenario analysis, and anything else that’s needed. Quit blind execution; focus instead on empowering leaders to make decisions.\n  get into the habit of understanding details, especially if something lies outside of your core competencies. If you’re dealing with data, or tech, or APIs, get under the hood.\n\n\nEven in cases where you don’t have a lot of decision-making due to organizational politics and dynamics, following the above recipe should earn you a lot of street cred. In your very localized environment, it will truly teach you how to make decisions."
    },
    
    {
      "type": "post",
      "title": "6 pro tips on communication that's sure to get you promoted",
      "slug": "6-pro-tips-on-communication-thats-sure-to-get-you-promoted",
      "url": "https://www.varunchoraria.com/6-pro-tips-on-communication-thats-sure-to-get-you-promoted/",
      "date": "2025-06-15",
      "tags": ["communication","career","leadership"],
      "excerpt": "I’ve been in the workforce for almost a decade now, and I’ve observed a rather persistent truth: being a great IC gets you noticed, but becoming a great communicator is what gets you promoted.",
      "content": "I’ve been in the workforce for almost a decade now, and I’ve observed a rather persistent truth: being a great IC gets you noticed, but becoming a great communicator is what gets you promoted.\n\nAs of writing this blog, I’ve been working with a large organization with 5,000+ employees. It’s a very dense talent pool, but after a point you get to see how some folks who were once your peers are now managing you or another team. Or even your peers who are getting high-impact and highly visible projects.\n\nIt’s not because they’re great at what they do, although sometimes that may be the case. They’re fantastic communicators. Communication is an extremely high-leverage skill to have, and no, it’s not only about writing pretty emails. These days even GPTs can do that, and a lot better than you too. So what is it?\n\n\n\n6 communication habits that compound\n\n\n  \n    Don’t save your best behavior for execs only: Getting promoted is a function of managing both up and across. We often only save the prettiest decks and the most compelling narratives for a larger meeting with execs. Problem is, you’re not socializing your ideas and getting buy-in from your peers on an ongoing basis. Doing it on an ongoing basis not only makes them feel involved, it builds a perception that you’re taking the lead and driving it. Treat every Slack update with your peers as a practice ground.\n  \n  \n    “You’re not strategic enough”: This was common feedback from various managers over the years. Took me a while to wrap my head around it, but when I did, I saw a simple truth: I over-estimated the amount of buy-in I had. To me, great communication meant technical proactiveness. “Enable” your leaders and every person cc’d with all the information, all at once. What this missed was the amount of work I was expecting someone else to do just to buy into an idea. Something I learned very recently was to sell your idea. It’s comms 101, where you wrap your idea into the 5 whys:\n\n    \n      “Why is this idea relevant now?”\n      “Why is this idea going to have X impact?”\n      “Why is this idea important for my team to pursue now?”\n      “Why will cross-functional teams prioritize this idea now?”\n      “Why are dollar and time resources for this idea justifiable now?”\n    \n\n    This doesn’t have to be a deck or a long doc. It can just be a few lines on Slack or email, thoughtfully convincing your leaders on the idea, what’s at stake if you don’t pursue it, and how it is aligned with team and company charter using only the resources already available.\n  \n  \n    Do not shoot yourself in the foot: I’ve probably done this 6 out of 10 times, because I was too enthusiastic. Leaders in meetings typically go “yes, yes, next, next” when you’ve spent the last week combing through reports to build a case. My only comms advice here: take the yes and move on. Do not double click on it unless they’ve explicitly raised an objection, or you’ll wind up convincing them out of the idea.\n  \n  \n    Prep like you’re taking the SATs: Often your pre-reads might be very detailed, but if you haven’t internalized the idea very well you will fumble and stammer, taking away credibility. Spend a lot of time editing, understanding, and going over the details. Find ways to trim it down to the absolute essentials. Read the room and think on your feet.\n  \n  \n    Simulate objections: Building on top of 4, I often use AI to poke holes in my idea. I upload the deck or doc, along with an ICP deck, and ask it to poke tactical holes. I sometimes describe what my colleagues and presenters are like from past conversations and the objections they’ve raised. This will help you think a couple of steps ahead in a very structured manner.\n  \n  \n    How confident should I be?: I’ve been in rooms where I was flushed with imposter syndrome but nevertheless had to present. I’ve also been among leaders I thought were the living embodiment of idiocy and paid the price with my over-confidence. If something is not backed by facts, data, or a case study, it’s a hunch, no matter how strong the conviction. Don’t mislead a leader. If something is a hunch or a hypothesis, say it.\n  \n\n\nStaying calm in high-stakes conversations\n\nOver and above these tips, the real world isn’t as black and white where you can easily apply X tip to Y scenario. Why? Emotions. We’re emotional beings, and so are our leaders.\n\nSo how do you stay calm in these high-stakes situations? Here is what I do:\n\n\n  when in doubt, always paraphrase the question asked and confirm your comprehension\n  if it’s a dicey 1:1, or if I’m getting into a call with someone I have a rocky working relationship with, I genuinely drive the conversation: enlist the bullets to discuss, seek alignment on the same, discuss items and note it, then send minutes\n  if I don’t know the subject very well and don’t have an answer to an objection, I note it down as an action item and give them a timeline of when I’ll be able to revert\n  if I know the subject well but don’t have an exact answer, I try to answer in the general direction and validate what they might really be asking\n\n\nSmall habits that build communication skill\n\nLastly, there are some ad-hoc skills you can pick up too, which compound your communication skills:\n\n\n  giving constructive feedback\n  putting everything on email, yes, despite the same things discussed on Slack\n  getting on calls and driving them, even if it’s a project where you’re just the contributor\n\n\nThat means taking initiative to:\n\n\n  send pre-reads\n  explain context and decision items within the call invite\n  spend the first 2 or 3 minutes discussing the objectives of the meeting\n  mediate it like you’re emceeing it\n  send minutes\n  follow up on minutes on Slack without cluttering threads"
    },
    
    {
      "type": "post",
      "title": "What your career path eventually comes down to",
      "slug": "what-your-career-path-eventually-comes-down-to",
      "url": "https://www.varunchoraria.com/what-your-career-path-eventually-comes-down-to/",
      "date": "2025-03-08",
      "tags": ["career","strategy"],
      "excerpt": "I’ve spent most of the last 6 years trying to find the answer. Along the way, I’ve seen a pretty clear difference between the low and the top performers.",
      "content": "I’ve spent most of the last 6 years trying to find the answer. Along the way, I’ve seen a pretty clear difference between the low and the top performers.\n\nLow performers\n\n\n  bootlick their way into most things\n  optimize for money, interviews, and gamifying career ladders\n  have no love for their craft\n  and the worst, have no curiosity\n\n\n\n\nTop performers\n\n\n  curiosity above and beyond what they’re doing\n  craft before paychecks\n  some of the hardest people to hire\n  some of the hardest people to command respect from\n  lone wolves\n  say no to opportunities if it doesn’t serve them, even if it comes from top management\n\n\n6 levels of career independence\n\nIf you want an infinite shelf life at work, or want to master pivots, or in general want to command premium pay, here’s what the career ladder to becoming a top performer comes down to:\n\n\n  you get told what to do and how to do it\n  you get told what to do\n  you get asked how to do it\n  you get asked what to do and how to do it\n  no one tells you anything\n  no one asks you anything\n\n\nThe sooner you can get to number 6, not for a paycheck or anything else, but to be worth your salt, the sooner you’ll start scratching the surface in any field. So, there’s a direct correlation between curiosity, hustle, and career success.\n\nHow to compound career durability\n\n\n  Don’t job hop for money. Gauge and bet on opportunities. Know when to exit.\n  Don’t job hop for title and salary. Arrogance has never gotten anyone far. Pick a job that has the best manager you can learn from.\n  Don’t job hop because of what your peers are doing.\n  Think 3 to 5 years ahead, minimum.\n  And last of all, really put in the work. You can almost always out-work anyone."
    },
    
    {
      "type": "post",
      "title": "Problems tell you what to fix. Pain points tell you why it matters.",
      "slug": "problems-tell-you-what-to-fix-pain-points-tell-you-why-it-matters",
      "url": "https://www.varunchoraria.com/problems-tell-you-what-to-fix-pain-points-tell-you-why-it-matters/",
      "date": "2025-01-30",
      "tags": ["product-marketing","sales"],
      "excerpt": "Every marketer, founder, or sales professional starts with the same question: “What’s the pain point of my prospect?” But too often, we conflate pain with problems, and that mistake costs us deals, loyalty, and growth.",
      "content": "Every marketer, founder, or sales professional starts with the same question: “What’s the pain point of my prospect?” But too often, we conflate pain with problems, and that mistake costs us deals, loyalty, and growth.\n\nLet’s break down the difference, why it matters, and how to leverage both to close more sales.\n\n\n\nThe critical difference between problems and pain\n\nAt first glance, problems and pains seem identical. They’re not.\n\n\n  A problem is a situational challenge.\n    \n      Example: “We need a CRM to track prospects.”\n    \n  \n  A pain is the emotional, psychological, or social consequence of that problem.\n    \n      Example: “If I don’t track prospects accurately, I’ll lose my job.”\n    \n  \n\n\nProblems are macro. Pains are micro. Problems are logical. Pains are human.\n\n\n\n\n\nThe science of pain: It’s not just physical\n\nPain is multidimensional. The bio-psycho-social model explains it as a mix of:\n\n\n  Biological factors (for example, physical strain from repetitive tasks)\n  Psychological factors (for example, stress or fear of failure)\n  Social factors (for example, losing status or missing promotions)\n\n\nIn business, pain is the intersection of a problem and a person. The same problem can cause wildly different pains for different stakeholders. Credit to Jonathan Stark’s article.\n\n\n\nWhy sales teams fail when they ignore pains\n\nFocusing only on solving problems leads to generic pitches. But when you address pains, you resonate emotionally. Let’s dissect an example.\n\nProblem: “We need a CRM.”\n\nPains vary by role:\n\n1. Sales Rep\n\n\n  Fear: “If I don’t hit quota, I’ll get fired.”\n  Frustration: “I waste hours on manual data entry.”\n  Emotional need: autonomy, security, recognition.\n\n\n2. Sales Manager\n\n\n  Fear: “If my team underperforms, I’ll lose credibility.”\n  Frustration: “I can’t identify coaching gaps without data.”\n  Emotional need: control, respect, career growth.\n\n\n3. Sales Leader\n\n\n  Fear: “Revenue leaks will make us miss targets.”\n  Frustration: “I can’t prioritize high-value accounts.”\n  Emotional need: strategic impact, reputation, scalability.\n\n\nIf your CRM pitch only highlights tracking prospects, you’ll miss the mark. But if you speak to the fear of failure for reps, the need for control for managers, and the desire for strategic wins for leaders, you’ll win buy-in at every level.\n\n\n\nHow to uncover pains, not just problems\n\n\n  Ask “why” repeatedly\n    \n      Problem: “We need better pipeline visibility.”\n      Pain: “Why?” -> “Because stalled deals hurt revenue.”\n      Deeper pain: “Why?” -> “If revenue drops, I’ll lose investor trust.”\n    \n  \n  Map pains to roles\n    \n      Create personas. A CFO’s pain, budget overruns, is not the same as a CMO’s pain, low lead quality.\n    \n  \n  Use emotional language\n    \n      Swap “increase efficiency” with “eliminate the anxiety of missed deadlines.”\n    \n  \n\n\nNext time you craft a pitch, ask: “Am I addressing the problem, or am I healing the pain?”"
    }
    
  ],
  "pages": [
    {
      "type": "page",
      "title": "About",
      "slug": "about",
      "url": "https://www.varunchoraria.com/about/",
      "intro": "B2B SaaS marketing operator. 8+ years of pipeline, positioning and product marketing, now built with agents.",
      "listing": false,
      "content": "I’m a marketing professional and I have worked in the B2B SaaS space for 8+ years now. I’m also a speaker, author, ex-host of a podcast, bodybuilder, and martial artist. I moonlight as a cat dad.\n\n\n\nNow\n\nI’m a B2B SaaS marketing operator. I help teams decide what marketing should do next, ship the work, and measure the result in pipeline, revenue, or another business outcome. I work across growth strategy, product marketing, and AI. Most recently, I led marketing at GTM Buddy (Series A) through August 2026, helping create the Revenue Activation category. The full history is here.\n\nI’m open to both full-time roles and consulting engagements. If you’re building something in AI or B2B and want help marketing it, here’s how I work.\n\nI’m all-in on Agentic AI. I believe the future will be one-person marketing teams, running a swarm of agents. GTM motions will be shipped in < 1 day, and scaled and managed with the help of agents. I’m building that way of working now.\n\n\n\nWhat I’ve done\n\nUni days\n\nI founded my first agency when I was 19, and worked with enterprise clients. During the 3 years that I ran the agency, we offered services like website, design, video, content and digital marketing.\n\nDuring that time, I also spoke at events across various universities, wrote a book, gave talks at various marketing conferences. I also built a no-code MBA bot on Facebook Messenger that taught the core frameworks of business in an interactive manner. It got 1K+ users in <24 hours.\n\nBy qualification, I’m a mechanical engineer. I was mostly an average student, scraping by in academics. However, my final year project was funded by the State Govt. of Karnataka. To this day, I don’t know why.\n\nEarly years\n\nI got my first job at a Series C startup, Vymo, and was the founding product marketing member. I helped position the product and enable sales as Vymo grew ARR 4x over 5 years. The top 5 banks and insurers across APAC were our customers. I saw the journey from Series A to Series C.\n\nI was the go-to guy for building best-in-class presentations, homegrown no-code tools, and being in lockstep with sales across deals.\n\nI also relocated to Chicago, Illinois to lead product marketing there. I helped crack our first $M deals during my time. Still a Bears fan to date.\n\nOn the side, I consulted and freelanced with multiple startups to help them build tools using Notion and Glide. I also supported them in running demand gen, sales enablement and product marketing.\n\nOne side quest I took up was running my CHRO podcast during the COVID years. I spoke to 30+ leaders from GoJek, Oracle, Infosys, MetLife, Razorpay and more.\n\nRecent past\n\nI joined Freshworks after Vymo. I took up the job because I wanted to explore something net new. I also wanted to see how things worked at scale. I picked up the PLG charter and set up the new business and expansion PLG motion from scratch, opening up a whole new revenue channel with $0 marketing spend.\n\nNot many side quests here, work kept me incredibly occupied. I did speak at another marketing conference about setting up PLG motions, and did some consulting work.\n\n\n\nWhat I’m into besides work\n\nI love science fiction, horror, mystery and psychological thrillers. I love watching films, and probably spend too much money on that aside from my app subscriptions every month.\n\nFor fun, I’m into bodybuilding. I used to be a swimmer, a long-distance cyclist and I also have a black belt in Tae Kwon Do. Sports and fitness are a crucial part of my everyday."
    },
    {
      "type": "page",
      "title": "ai",
      "slug": "tags/ai",
      "url": "https://www.varunchoraria.com/tags/ai/",
      "intro": "I write about AI agents, human judgment, and practical uses of machine intelligence at work.",
      "listing": true,
      "content": "I write about AI agents, human judgment, and practical uses of machine intelligence at work.\n- My agent found 9.09%. I found one session. (2026-08-03) https://www.varunchoraria.com/11-sessions-are-not-a-ux-audit/: My agent found a UX problem on my site. Dead clicks had affected 9.09% of sessions, above the 5% threshold. It filed a GitHub issue and told me what to inspect.\n- The models were fine. I was the bug. (2026-08-02) https://www.varunchoraria.com/the-models-were-fine-i-was-the-bug/: I typed a command and nothing happened. No output, no error, no prompt coming back, just a cursor sitting there while my laptop got warm.\n- The Claude bill said $1,492. I paid $20. Here's how. (2026-07-30) https://www.varunchoraria.com/how-to-get-1492-out-of-a-20-claude-subscription/: I pay $20 a month for Claude. Last week I went digging through my own logs to find out what that actually buys, because I wanted to write an honest post about it and I did not want to guess.\n- Hunting Season for the Rest of Us (2026-07-24) https://www.varunchoraria.com/hunting-season-for-the-rest-of-us/: About a week into prepping for one of these companies, I could drop the category’s favourite piece of jargon into a sentence without flinching. Felt good about it too.\n- Github-ing (2026-07-16) https://www.varunchoraria.com/github-ing/: The world is moving toward systems-first work, and I do not think most people understand what that means yet.\n- How to design without AI Slop (2026-07-08) https://www.varunchoraria.com/how-to-design-without-ai-slop/: You are what you admire\n- I now have my own Master Shifu (2026-07-02) https://www.varunchoraria.com/i-now-have-my-own-master-shifu/: I’ve been collecting consulting casebooks for about a decade. Kellogg, Wharton, Ross, Darden, IIM (A, B, C, I, L), MDI, Drucker, McKinsey, Victor Cheng, Crack the Case — 19 books in total, some of them 300+ pages.\n- Nobody uses my MCP server. Building it still worked. (2026-06-28) https://www.varunchoraria.com/nobody-uses-my-mcp-server-building-it-still-worked/: Nobody’s going to read this, and I don’t mean this post specifically. I mean the whole thing: the blog, the about page, the work history. A hiring manager has 400 applications open and 11 minutes. A reader lands on one essay from a link ...\n- Killed By Google (2026-06-16) https://www.varunchoraria.com/killed-by-google/: I first came across killed by google and was honestly quite blown. I suppose the startup myth is real: your instinct is right 95% of the time, but your ideas are wrong probably 80% of the time.\n- Who owns what? (2026-05-17) https://www.varunchoraria.com/who-owns-what/: Everyone these days is asking the same question in the age of agentic AI: what should an AI do vs. what should a human do?\n- Imposter syndrome as a first-time manager (2026-04-12) https://www.varunchoraria.com/imposter-syndrome-as-a-first-time-manager/: It’s been about a quarter since I became a manager. I’m wondering what value I’m actually adding to the organization, especially in the agentic era where skills, information, and knowledge are more or less commoditized. Here are my candi..."
    },
    {
      "type": "page",
      "title": "analytics",
      "slug": "tags/analytics",
      "url": "https://www.varunchoraria.com/tags/analytics/",
      "intro": "I write about measurement, evidence, observability, and turning data into useful decisions.",
      "listing": true,
      "content": "I write about measurement, evidence, observability, and turning data into useful decisions.\n- My agent found 9.09%. I found one session. (2026-08-03) https://www.varunchoraria.com/11-sessions-are-not-a-ux-audit/: My agent found a UX problem on my site. Dead clicks had affected 9.09% of sessions, above the 5% threshold. It filed a GitHub issue and told me what to inspect.\n- The Claude bill said $1,492. I paid $20. Here's how. (2026-07-30) https://www.varunchoraria.com/how-to-get-1492-out-of-a-20-claude-subscription/: I pay $20 a month for Claude. Last week I went digging through my own logs to find out what that actually buys, because I wanted to write an honest post about it and I did not want to guess.\n- Maybe going around CIRCLES is worth it (2026-06-26) https://www.varunchoraria.com/maybe-going-around-circles-is-worth-it/: Frameworks get a bad rep. But early in your problem-solving journey, they’re scaffolding — not a crutch. Over time, as you get acquainted with more real-life scenarios, your intuition gets powerful enough to spot-check and work bottoms-up."
    },
    {
      "type": "page",
      "title": "career",
      "slug": "tags/career",
      "url": "https://www.varunchoraria.com/tags/career/",
      "intro": "I write about career durability, professional growth, luck, craft, and navigating modern work.",
      "listing": true,
      "content": "I write about career durability, professional growth, luck, craft, and navigating modern work.\n- Hunting Season for the Rest of Us (2026-07-24) https://www.varunchoraria.com/hunting-season-for-the-rest-of-us/: About a week into prepping for one of these companies, I could drop the category’s favourite piece of jargon into a sentence without flinching. Felt good about it too.\n- Github-ing (2026-07-16) https://www.varunchoraria.com/github-ing/: The world is moving toward systems-first work, and I do not think most people understand what that means yet.\n- Nobody uses my MCP server. Building it still worked. (2026-06-28) https://www.varunchoraria.com/nobody-uses-my-mcp-server-building-it-still-worked/: Nobody’s going to read this, and I don’t mean this post specifically. I mean the whole thing: the blog, the about page, the work history. A hiring manager has 400 applications open and 11 minutes. A reader lands on one essay from a link ...\n- Killed By Google (2026-06-16) https://www.varunchoraria.com/killed-by-google/: I first came across killed by google and was honestly quite blown. I suppose the startup myth is real: your instinct is right 95% of the time, but your ideas are wrong probably 80% of the time.\n- I've been trying to get lucky for a decade. Here's what I learnt. (2026-04-19) https://www.varunchoraria.com/ive-been-trying-to-get-lucky-for-a-decade/: I think one of the things I have a perpetual obsession with is career optimization. It feels a lot like poker. There is no magic playbook, or a clean meritocratic path to it, at least not anymore.\n- Imposter syndrome as a first-time manager (2026-04-12) https://www.varunchoraria.com/imposter-syndrome-as-a-first-time-manager/: It’s been about a quarter since I became a manager. I’m wondering what value I’m actually adding to the organization, especially in the agentic era where skills, information, and knowledge are more or less commoditized. Here are my candi...\n- 6 pro tips on communication that's sure to get you promoted (2025-06-15) https://www.varunchoraria.com/6-pro-tips-on-communication-thats-sure-to-get-you-promoted/: I’ve been in the workforce for almost a decade now, and I’ve observed a rather persistent truth: being a great IC gets you noticed, but becoming a great communicator is what gets you promoted.\n- What your career path eventually comes down to (2025-03-08) https://www.varunchoraria.com/what-your-career-path-eventually-comes-down-to/: I’ve spent most of the last 6 years trying to find the answer. Along the way, I’ve seen a pretty clear difference between the low and the top performers."
    },
    {
      "type": "page",
      "title": "Changelog",
      "slug": "changelog",
      "url": "https://www.varunchoraria.com/changelog/",
      "intro": "Only changes worth noticing: what became easier to find, understand, use, or trust, and why I changed it.",
      "listing": false,
      "content": "This updates with the site. Releases worth noting carry structured visitor notes in Git history. Deployment validates and publishes those notes, skips internal work, and needs no external model. This stays current without becoming a raw commit dump.\n  \n  Last updated 9 August 2026.\n\n\n\nThis is not a mirror of the Git history. A renamed file is not an update. Neither is a cleaner build script, unless it changes something you can see, use, find, or trust.\n\nThe entries below keep the consequence and drop the maintenance log.\n\n\n  \n    newer\n    \n      Month and year\n      \n    \n    older\n  \n  \n  \n  \n    \n      9-aug-2026\n      The site now follows your system dark mode\n      If your device is set to dark, this site is now dark too. There is no toggle to find and nothing is stored about you: the page reads your operating system setting and matches it. Body text sits at 14:1 contrast against a ground that is deliberately not pure black, because pure black under near-white text makes letters glow at the edges and is hardest on readers with astigmatism.\n      \n      \n        Reading at night: Text, links, dates and quotes keep the same order of emphasis they have in light mode, so nothing shouts that did not shout before.\n      \n        No setting to find: Your device already knows which you prefer, so the site just uses that. Nothing to click, and no preference kept about you.\n      \n        Contribution graph: The green scale switches to GitHub's dark values, so a busy day still reads brighter than a quiet one instead of the reverse.\n      \n      \n    \n  \n    \n      3-aug-2026\n      The first 30 days page says what it means\n      The page describing what happens in month one of an engagement was written in vague, passive language that named categories instead of things. It now uses plain sentences and lists, and states plainly that in two of four engagements the finding was that the problem was not marketing.\n      \n      \n        Easier to skim: The metrics, targeting changes and day-30 deliverables are lists now, not paragraphs to read through.\n      \n        Clearer answers: The questions at the bottom match what people actually ask, so both readers and AI assistants get a direct answer.\n      \n      \n    \n  \n    \n      3-aug-2026\n      Pages inside consulting now show where they sit\n      A page filed under consulting was showing as though it sat directly under the home page, both in the trail at the top and in the structured data search engines read. It now shows the full path, and the first 30 days page is linked from the footer.\n      \n      \n        Knowing where you are: The trail at the top of a consulting page now names the section it belongs to instead of skipping it.\n      \n        Correct in search: Search engines read the same corrected path, so results can show the section rather than a page floating on its own.\n      \n      \n    \n  \n    \n      3-aug-2026\n      The consulting work now names its clients and shows the first month\n      Four consulting engagements that were listed only by country and funding stage are now named, with what the work was in each. A new page walks through what actually happens in the first 30 days of an engagement, drawn from those same four rather than from a generic process.\n      \n      \n        Checkable claims: Named clients can be verified. A geography and a funding stage cannot.\n      \n        Fewer questions before a call: What the first month involves is now readable in advance instead of being the first thing a call has to cover.\n      \n        Honest about the order: The page says the audit comes before the strategy, including where that meant telling a client their problem was not marketing.\n      \n      \n    \n  \n    \n      3-aug-2026\n      The uses page now shows what each tool is, and links to it\n      Every tool on /uses-this/ now carries its own logo and its own link, so you can see and reach the thing being described instead of reading a name. The tools I ran at my last job moved into a separate section, because a page called uses this should not quietly claim a stack I no longer pay for.\n      \n      \n        Faster scanning: Each row leads with the tool's logo, so you can find the one you came for without reading every line.\n      \n        Every tool is reachable: All 39 entries link to the product, including each name inside rows that list several.\n      \n        Honest about the stack: Tools from my last role are listed as tools I have run, not tools I use today.\n      \n      \n    \n  \n    \n      3-aug-2026\n      The homepage now gives visitors three plain next steps\n      Hiring managers, consulting buyers, and people ready to talk no longer have to unpack one crowded paragraph. Each path now gets its own short line and direct link, without phrases like full record or how engagements run.\n      \n      \n        Faster choice: Visitors can choose full-time work, consulting, or a call without reading a combined pitch.\n      \n        More human copy: Generic link labels were replaced with direct language that matches the rest of the site.\n      \n      \n    \n  \n    \n      3-aug-2026\n      The 9.09% essay now follows the actual mistake\n      The essay no longer mixes a small sample, a duplicate issue bug, and page-level analytics into competing stories. It now follows one clear mistake from a single dead-click session to an unjustified UX recommendation, with a sharper title to match.\n      \n      \n        Clear argument: The post now explains exactly how a correct percentage became a bad recommendation.\n      \n        Better title: The new title uses the same concrete contrast as the site's older essays.\n      \n      \n    \n  \n    \n      3-aug-2026\n      Changelog navigation now looks like the rest of the site\n      The boxed month filter is gone. Changelog browsing now uses plain underlined controls, while thin rules separate navigation and every update. The same month selection, URL state, keyboard behavior, and no-JavaScript fallback remain intact without introducing a separate visual language.\n      \n      \n        Simpler controls: Month browsing now looks like ordinary site navigation instead of a form panel.\n      \n        Clear separation: Thin rules make each update easier to distinguish while scanning.\n      \n        Mobile fit: The underlined controls stay on one row, retain 44px targets, and fit at 320px.\n      \n      \n    \n  \n    \n      3-aug-2026\n      Changelog now stays current and browses cleanly by month\n      The stale automation no longer depends on retired GitHub Models. Verified commit metadata now updates the public record during deployment, while month-and-year controls let readers move through releases without scrolling through the complete chronology.\n      \n      \n        Reliable updates: Deployment no longer silently keeps old data when a model endpoint disappears.\n      \n        Month browsing: Newer, older, and month-and-year controls make long history easier to scan.\n      \n        Mobile fit: Controls use 44px targets, fit at 320px, and preserve the full list when JavaScript is unavailable.\n      \n      \n    \n  \n    \n      2-aug-2026\n      Work pages now put each claim in the right place\n      The homepage now leads with team scope and category creation at GTM Buddy, while the full Work page carries sourced pipeline and budget details. Freshworks keeps the absolute ARR private and states $0 in paid media. Vymo’s 4x ARR growth is clearly company context rather than a personal revenue claim.\n      \n      \n        Cleaner attribution: Pipeline is labelled sourced, while company growth is no longer written like one person caused it.\n      \n        Public boundaries: Freshworks keeps the absolute figure off the site without hiding the outcome.\n      \n        Right level: The homepage stays concise, while Work carries the detailed operating record.\n      \n      \n    \n  \n    \n      2-aug-2026\n      The 9.09% essay now explains what the denominator means\n      The essay about Lazarus Pit no longer treats one changed recommendation out of eleven like a grand success rate. It now says plainly that eleven sessions are too small for precision, links the tool being discussed, and keeps the useful point: disagreement between agents exposed a flaw in the review loop.\n      \n      \n        Clear denominator: The essay explains the sample behind 9.09% instead of asking the percentage to sound meaningful.\n      \n        Useful context: Lazarus Pit is linked where the experiment first appears.\n      \n        Human voice: The rewrite drops decorative phrasing and states what the experiment did and did not prove.\n      \n      \n    \n  \n    \n      1-aug-2026\n      Rainmaker setup now starts with one command and informed consent\n      Rainmaker’s page now leads with a skill-first setup, points installation at the npm package, and explains when provider access costs money or uses quota. Readers can install the tool without reverse-engineering the repository, bring their own model, and know that paid or quota-backed providers will not be activated without consent.\n      \n      \n        Start faster: The documented install path begins with one command instead of repository archaeology.\n      \n        Choose providers: The guide separates the core workflow from optional model and crawl providers.\n      \n        Control spend: Paid or quota-backed services require explicit consent before activation.\n      \n      \n    \n  \n    \n      1-aug-2026\n      Code blocks now copy in one tap\n      Code examples now get a small Copy button without changing the underlying text or requiring a framework. The control stays visible on touch devices, gives keyboard users a clear focus target, and reports when copying succeeds. Selecting the code manually still works when JavaScript or clipboard access is unavailable.\n      \n      \n        Copy faster: Commands and examples can move to the clipboard with one tap.\n      \n        Works on touch: The control remains visible on phones where hover does not exist.\n      \n        Keeps fallback: Code stays selectable when scripting or clipboard access is unavailable.\n      \n      \n    \n  \n    \n      30-jul-2026\n      Rainmaker now clarifies its role as both SEO and AEO agent\n      The Rainmaker side project page and listings now describe the tool as an open-source SEO and AEO agent, not just a search system. This change clarifies its purpose for visitors seeking agents that handle both search engine and answer engine optimization, preventing confusion about its scope and use cases.\n      \n      \n        Clear positioning: Rainmaker’s page and directory entry now state it is for both SEO and AEO, so visitors know what problems it solves.\n      \n        Find the right tool: The updated description helps visitors searching for answer engine optimization agents discover Rainmaker without guessing.\n      \n        Accurate summaries: Meta descriptions and intros now match the tool’s actual job, reducing ambiguity for new readers.\n      \n      \n    \n  \n    \n      30-jul-2026\n      Rainmaker skill tree diagram now shows the missing Ground phase\n      The Rainmaker side project page updates its skill tree diagram to include the Ground phase, which was previously absent. The diagram and description now clarify all six phases and their roles, so visitors can understand how context is written and used in the system. This prevents confusion about the skill tree’s structure and sequence.\n      \n      \n        Understand system flow: The diagram now shows how the Ground phase writes context for other skills.\n      \n        Find missing phase: Visitors can see all six phases, not just five, and follow the job sequence accurately.\n      \n        Clearer onboarding: The skill tree explanation matches the actual implementation, reducing onboarding guesswork.\n      \n      \n    \n  \n    \n      30-jul-2026\n      Rainmaker side project now public with full landing page\n      Rainmaker, an open-source search system that ranks findings by distance to revenue, now has a public landing page and repository. The page explains the job to be done, onboarding steps, skill tree, and how Rainmaker differs from typical SEO tools. Visitors can see diagrams, example prompts, and clear answers about usage, licensing, and requirements.\n      \n      \n        Understand the tool: The landing page explains what Rainmaker is for, how it works, and who should use it.\n      \n        Find the repository: A public GitHub link lets visitors access the code, specification, and license directly.\n      \n        See onboarding steps: Diagrams and prompts clarify the workflow, skill tree, and how Rainmaker fits different site types.\n      \n      \n    \n  \n    \n      30-jul-2026\n      Letterboxd icon now recognizable in monochrome\n      The Letterboxd social icon on the homepage now uses its rounded badge form, making the logo readable at small sizes without relying on color. This change prevents the icon from appearing as three generic dots and matches the handling of other social icons. Visitors can identify and access the Letterboxd profile more easily.\n      \n      \n        Clear iconography: The Letterboxd logo is visually distinct and recognizable even in monochrome.\n      \n        Consistent navigation: Social icons now follow a uniform style, helping visitors find profiles without confusion.\n      \n        Accessible design: The icon remains legible at small sizes, supporting users on various devices.\n      \n      \n    \n  \n    \n      30-jul-2026\n      Cursor profile now visible among homepage social icons\n      The homepage adds a Cursor.com profile link alongside LinkedIn, Twitter, GitHub, and Letterboxd, using a matching icon and accessible label. This makes the full set of public profiles discoverable from the main page, so visitors can find and connect through Cursor without guessing or hunting for the handle.\n      \n      \n        Find all profiles: Cursor joins the homepage social row, so every public profile is visible in one place.\n      \n        Accessible navigation: Each icon uses a clear label and title, helping screen readers and keyboard users.\n      \n        Consistent design: The icon row stays uniform, with no extra boxes or external-arrow marks.\n      \n      \n    \n  \n    \n      29-jul-2026\n      New pages now auto-appear in MCP feed for AI access\n      Pages with titles and directory-style URLs are now included in the MCP feed by default, unless intentionally hidden or flagged out. This removes the need to manually add a flag for every new page, making AI-powered tools and clients reliably see fresh content without extra setup. Posts remain automatically included.\n      \n      \n        Find new pages: AI tools and clients can read new pages as soon as they are published, without waiting for manual flags.\n      \n        Less setup friction: Site editors no longer need to remember or add a special flag for every new page.\n      \n        Trust feed completeness: The feed stays in sync with the sitemap and build checks, so visitors can trust that all public content is accessible.\n      \n      \n    \n  \n    \n      29-jul-2026\n      MCP page now explains how the server stays current\n      The MCP section adds a clear explanation of how its server stays up to date: it reads a build-time feed refreshed every five minutes, and deploys fail if any public page is missing from that feed. Tool descriptions are also corrected to match the current site structure and search behavior. This helps visitors trust the freshness and completeness of AI-powered access.\n      \n      \n        Understand freshness: Visitors can see exactly how the server stays current and why new pages cannot quietly go unreadable.\n      \n        Trust AI access: The mechanism is explained, so readers know agents see what humans see, with no hidden gaps.\n      \n        Correct tool info: Descriptions now match the actual site pages and search ranking, avoiding confusion from outdated labels.\n      \n      \n    \n  \n    \n      29-jul-2026\n      Listing pages in MCP feed now flagged for search weighting\n      The site’s API feed now marks homepage, blog, and tag archive listing pages with a 'listing' flag. This helps clients and tools distinguish between generated listings and original essays, so full-text search and ranking can treat them appropriately. The change prevents listings from outranking actual content based on excerpt matches alone.\n      \n      \n        Clearer search results: Listings and essays are now differentiated, so search tools can rank pages more accurately.\n      \n        Machine-readable structure: API clients can detect listing pages and adjust their behavior without guessing from content.\n      \n      \n    \n  \n    \n      29-jul-2026\n      Consulting now explains solo, AI-native work under one page\n      The consulting section now merges all previous service pages into a single, clearly positioned offering. The site drops references to Grow & Close and clarifies that consulting is solo and AI-native, with strategy and execution handled by one person. This helps visitors understand the engagement shape and find the right page without guessing between options.\n      \n      \n        Find the right page: Consulting, product marketing, and AI marketing now live under one URL with redirects from the old pages.\n      \n        Clear engagement shape: The site explains that consulting is solo and agentic, not an agency or studio.\n      \n        Footer stays simple: Footer links now point only to consulting, work, and booking a call, reducing confusion.\n      \n      \n    \n  \n    \n      29-jul-2026\n      MCP feed now exposes readable page slugs and listing content\n      The site’s API feed now uses readable page slugs and names the homepage 'home'. Listing pages like blog and tag archives now expose their intros and post lists, so AI clients and tools can read and summarize them. This prevents empty content and makes navigation and discovery easier for machine readers.\n      \n      \n        Machine-readable navigation: Home, blog, and tag pages now have clear slugs and content for AI clients.\n      \n        Listing content included: Listing pages expose intros and post summaries instead of empty bodies.\n      \n      \n    \n  \n    \n      29-jul-2026\n      Consultation booking intent tracked as a key event\n      Booking a consultation via Cal.com now triggers a stable custom event in Google Analytics. This lets site owners reliably measure consultation intent and treat it as a key event, improving insight into visitor actions without affecting site performance or usability.\n      \n      \n        Reliable analytics: Cal.com booking clicks are tracked with a dedicated event name for consultation intent.\n      \n        QA coverage: Site QA checks that booking events are emitted, preventing silent analytics failures.\n      \n      \n    \n  \n    \n      28-jul-2026\n      Private side quests now have public landing pages and clear labels\n      Lazarus Pit, previously undocumented, now has a landing page explaining its self-healing UX agent, pipeline, and public issue output. All side quests are labelled Public or Private everywhere, so visitors can tell which projects are accessible and which are documented but not open source. Private quests link only to on-site pages, never to inaccessible repositories.\n      \n      \n        Find hidden projects: Lazarus Pit is now documented with a clear explanation and public output links.\n      \n        Know project access: Every side quest is labelled Public or Private on both the homepage and side-quests page.\n      \n        Avoid dead links: Private projects link only to their landing pages, not to inaccessible repositories.\n      \n      \n    \n  \n    \n      27-jul-2026\n      Footer links now use plain language and cover privacy, contact, and disclaimers\n      The footer now uses descriptive labels for tools, gear, and side projects, replacing jargon and project names with clear explanations. It also adds direct links to privacy, contact, and disclaimer pages, so visitors can find essential information and reach out without hunting through menus or guessing terminology.\n      \n      \n        Find essentials fast: Privacy, contact, and disclaimer links are now visible in the footer.\n      \n        Clear tool names: Side projects and gear are labeled in plain English, not just project titles.\n      \n      \n    \n  \n    \n      27-jul-2026\n      Images now preload for faster page loads on key pages\n      The site now preloads hero and featured images on the homepage and other important pages. This reduces the time visitors wait for visuals to appear, especially on slower connections or when landing directly on a major section. The change helps the site feel more responsive without adding complexity.\n      \n      \n        Quicker visuals: Hero and featured images load sooner, so pages appear complete faster.\n      \n        Better first impression: Landing pages no longer show empty spaces while images download.\n      \n      \n    \n  \n    \n      27-jul-2026\n      Visitor changelog now updates itself after every deploy\n      The site now synthesizes and publishes a visitor-facing changelog automatically after each deployment. This removes manual steps and ensures that site updates are recorded promptly and accurately, so visitors can trust the changelog to reflect what actually changed and why it matters.\n      \n      \n        Changelog stays fresh: Site updates are recorded immediately after each deploy, reducing stale or missing entries.\n      \n        Trustable history: Visitors can rely on the changelog for a clear, accurate record of changes that affect their experience.\n      \n        Less manual friction: Maintainers no longer need to manually update the changelog, lowering the risk of omissions.\n      \n      \n    \n  \n    \n      27-jul-2026\n      Side projects now explain the payoff before sending you to the code\n      Interview Recon, Master Shifu, and Michealangelo now have proper pages. You can work out what each tool is for, what it gives you, where it stops, and how to install it without reverse-engineering a README.\n      \n      \n        Choose faster: Each page leads with the outcome, then explains the mechanism and limits.\n      \n        Install without guesswork: Commands, supported agents, free-tier constraints, and expected outputs live in one place.\n      \n        See useful changes: Project updates ignore documentation noise and keep only product work that changes what the tool can do.\n      \n      \n    \n  \n    \n      25-jul-2026\n      Speaking and media stopped hiding behind a page called Fun\n      The old Fun page mixed a book, talks, and podcast appearances under a label that explained none of them. It is now Speaking & Media, with a useful URL and a short list of episodes worth starting with.\n      \n      \n        Find the proof: Talks, podcast seasons, and the book now sit under a name people would search for.\n      \n        Start somewhere useful: Eight recommended episodes replace the job of digging through two complete seasons.\n      \n        Keep old links working: The previous URL redirects instead of becoming a dead end.\n      \n      \n    \n  \n    \n      24-jul-2026\n      You can now tell what hiring me would involve before booking a call\n      The consulting section now separates general B2B SaaS work, AI marketing work, and product marketing work. Each page names the problem, the engagement shape, the expected outcome, and the situations where I am the wrong fit.\n      \n      \n        Evaluate fit: The pages explain workshops, defined projects, ongoing consulting, and fractional leadership without pretending every company needs the same thing.\n      \n        See the boundary: What I do not do is stated beside what I do.\n      \n        Reach the right page: Navigation and internal links now connect the broad consulting offer to its specialist pages.\n      \n      \n    \n  \n    \n      24-jul-2026\n      Shared links finally look like the page they lead to\n      Essays now generate wide preview images that survive LinkedIn, WhatsApp, and Slack. Pages also gained visible breadcrumbs, so arriving deep inside the site no longer means guessing where you are.\n      \n      \n        Know before clicking: Link previews carry the real title, description, and a readable image.\n      \n        Move around: Breadcrumbs show the page's place in the site and provide a short route back.\n      \n        Keep renamed writing reachable: Old essay URLs redirect instead of breaking shared links.\n      \n      \n    \n  \n    \n      18-jul-2026\n      Search engines and AI systems no longer have to guess what each page is\n      Titles, descriptions, canonical URLs, structured data, author details, heading order, RSS discovery, and crawl rules now describe the site consistently. Most of this is invisible until it is missing, which it was.\n      \n      \n        Better search results: Pages have distinct titles and descriptions instead of borrowing generic site copy.\n      \n        Cleaner indexing: Duplicate archive aliases and the 404 page no longer compete with real pages.\n      \n        Clearer machine context: Search engines and AI agents can identify the author, page type, canonical URL, and relationships between pages.\n      \n      \n    \n  \n    \n      8-jul-2026\n      The site now gives both humans and AI a sensible next step\n      Every page ends with direct routes to ask ChatGPT, Claude, or Perplexity about me, plus the site keeps RSS visible for people who would rather follow the writing without an algorithm.\n      \n      \n        Continue the question: AI links open with useful context instead of dropping you onto an empty chat.\n      \n        Follow the writing: RSS is linked from the site instead of existing as an undiscoverable feed.\n      \n        Keep the interface quiet: The calls to action use the same restrained visual system as the rest of the publication.\n      \n      \n    \n  \n    \n      7-jul-2026\n      The site became a publication instead of a pile of page templates\n      Every major page moved onto one narrow, readable system with shared type, spacing, navigation, and content patterns. Half-built features came out when they added maintenance without helping a reader.\n      \n      \n        Read without fighting the layout: Long-form pages, lists, tags, projects, and utility pages now share one predictable rhythm.\n      \n        Use it on small screens: Navigation, timelines, headers, and footers reflow cleanly down to phone widths.\n      \n        No pretend features: The unfinished email form and an intrusive MCP prompt were removed instead of being left to imply they worked.\n      \n      \n    \n  \n    \n      28-jun-2026\n      AI clients can read the live site instead of relying on stale training data\n      The MCP endpoint exposes current posts and pages to compatible AI clients. A pre-push QA gate also checks public content for search, answer-engine, design, and privacy mistakes before it leaves the repository.\n      \n      \n        Use current context: Compatible clients can search and retrieve the actual live publication.\n      \n        Connect with one URL: The MCP page gives client-specific setup snippets instead of requiring a local bridge.\n      \n        Reduce public mistakes: The QA gate blocks sensitive exports and catches broken content structure before push.\n      \n      \n    \n  \n    \n      23-jun-2026\n      The site became easier to find, faster to load, and less awkward to use\n      Search and AI crawl files were added, accessibility failures were fixed, and a tiny avatar stopped shipping as a 104KB image. None of that needed a redesign. It needed paying attention.\n      \n      \n        Readable controls: Contrast and screen-reader labels now match what people see.\n      \n        Less wasted data: The small booking image became roughly 37 times lighter at the same display size.\n      \n        Better discovery: robots.txt and llms.txt give search engines and AI systems explicit routes through the site.\n      \n      \n    \n  \n    \n      17-may-2026\n      The default Jekyll theme came out; a real personal site went in\n      I rebuilt the site around my writing, work, and side projects, then connected GitHub Pages so changes deploy from the repository. This is where the current site starts.\n      \n      \n        One home: Writing, work, speaking, and experiments became parts of the same site.\n      \n        Changes stay accountable: The public repository records what changed and GitHub Pages ships it."
    },
    {
      "type": "page",
      "title": "communication",
      "slug": "tags/communication",
      "url": "https://www.varunchoraria.com/tags/communication/",
      "intro": "I write about clear communication, earning trust, managing objections, and influencing decisions.",
      "listing": true,
      "content": "I write about clear communication, earning trust, managing objections, and influencing decisions.\n- 6 pro tips on communication that's sure to get you promoted (2025-06-15) https://www.varunchoraria.com/6-pro-tips-on-communication-thats-sure-to-get-you-promoted/: I’ve been in the workforce for almost a decade now, and I’ve observed a rather persistent truth: being a great IC gets you noticed, but becoming a great communicator is what gets you promoted."
    },
    {
      "type": "page",
      "title": "Contact",
      "slug": "contact",
      "url": "https://www.varunchoraria.com/contact/",
      "intro": "The shortest routes to a useful conversation.",
      "listing": false,
      "content": "Work together\n\nYou do not need to diagnose the problem before the call. Bring the pipeline model, positioning problem, launch plan, AI question, team problem, or list of competing priorities. We will work out whether it needs a workshop, a defined project, or an ongoing role.\n\nBook 30 minutes on my calendar. Want the detail first? The consulting page covers how I work and what I measure.\n\nWhat happens on the call\n\n30 minutes, no pitch. I will ask about your target, your buyers, and what has changed. You leave with a clearer next step even if the answer is that we should not work together.\n\nWho this is for\n\nB2B SaaS founders, marketing leaders, and teams that need senior marketing judgment without a full-time hire or an agency layer. The pattern is usually one of these:\n\n\n  The pipeline target and the marketing plan do not add up\n  The product changed, but the homepage still tells the old story\n  The team is busy, but the work does not build on itself\n  The company needs senior judgment, but not another full-time leader\n\n\nIf you recognize one of these, the call is worth 30 minutes. If you do not, the call will make that clear quickly, and neither of us loses much time.\n\nEvidence before a pitch\n\nThe tools behind this site are open source on GitHub. Read them, fork them, and decide with evidence instead of a pitch. I will not promise a rank, a vanity metric, or a launch headline in exchange for a contract.\n\nRoles, speaking, and everything else\n\nLinkedIn is the best place for full-time roles, speaking invitations, media requests, corrections, and questions about this site.\n\nFor a bug or idea tied to an open-source project, use the repository linked from that side quest. That keeps the context beside the code and makes it easier for me to act on."
    },
    {
      "type": "page",
      "title": "design",
      "slug": "tags/design",
      "url": "https://www.varunchoraria.com/tags/design/",
      "intro": "I write about useful interfaces, design systems, taste, and avoiding generic AI-generated work.",
      "listing": true,
      "content": "I write about useful interfaces, design systems, taste, and avoiding generic AI-generated work.\n- My agent found 9.09%. I found one session. (2026-08-03) https://www.varunchoraria.com/11-sessions-are-not-a-ux-audit/: My agent found a UX problem on my site. Dead clicks had affected 9.09% of sessions, above the 5% threshold. It filed a GitHub issue and told me what to inspect.\n- How to design without AI Slop (2026-07-08) https://www.varunchoraria.com/how-to-design-without-ai-slop/: You are what you admire"
    },
    {
      "type": "page",
      "title": "Disclaimer",
      "slug": "disclaimer",
      "url": "https://www.varunchoraria.com/disclaimer/",
      "intro": "Because the world takes things too literally these days",
      "listing": false,
      "content": "Website Disclaimer\n\nLast Updated: June 25, 2026\n\nGeneral Disclaimer\n\nThe information provided on this website is for general informational purposes only. All content on this website is provided in good faith, however, I make no representation or warranty of any kind, express or implied, regarding the accuracy, adequacy, validity, reliability, availability, or completeness of any information on this website.\n\nPersonal Views\n\nThe views, thoughts, and opinions expressed on this website are solely my own and do not necessarily reflect the views, opinions, positions, or strategies of my past, current, or future employers, business partners, clients, or any organizations I may be affiliated with. Any content, views, or opinions published on this website are my personal views and should not be attributed to any entity with whom I have been, am now, or will be affiliated.\n\nNo Professional Advice\n\nThe content on this website is not intended to be a substitute for professional advice. Before taking any actions based on the information provided on this website, please consult with appropriate professionals. The use or reliance of any information contained on this website is solely at your own risk.\n\nExternal Links\n\nThis website may contain links to external websites that are not provided or maintained by me. I do not guarantee the accuracy, relevance, timeliness, or completeness of any information on these external websites.\n\nUpdates and Changes\n\nI reserve the right to modify this disclaimer at any time without notice. By using this website, you agree to the current version of this disclaimer.\n\nAnalytics & Tracking\n\nThis website uses Microsoft Clarity to understand how visitors use and interact with it — including behavioral metrics, heatmaps, and session replays. This data is used solely to improve the site. Website usage data is captured using first and third-party cookies and other tracking technologies. By using this site, you agree to data collection by me and Microsoft. For more information on how Microsoft collects and uses your data, visit the Microsoft Privacy Statement.\n\nContact Information\n\nIf you have questions about this disclaimer, use the options on the contact page. For details about analytics, cookies, and the services that process site data, read the privacy page."
    },
    {
      "type": "page",
      "title": "frameworks",
      "slug": "tags/frameworks",
      "url": "https://www.varunchoraria.com/tags/frameworks/",
      "intro": "I write about working frameworks for strategy, management, prioritization, analysis, and better decisions.",
      "listing": true,
      "content": "I write about working frameworks for strategy, management, prioritization, analysis, and better decisions.\n- The models were fine. I was the bug. (2026-08-02) https://www.varunchoraria.com/the-models-were-fine-i-was-the-bug/: I typed a command and nothing happened. No output, no error, no prompt coming back, just a cursor sitting there while my laptop got warm.\n- Who do we sell to vs. Who SHOULD we sell to (and how). (2026-08-01) https://www.varunchoraria.com/who-do-we-sell-to-vs-who-should-we-sell-to-and-how/: I’ve been a part of both startups and large organizations, and one consistent trend I’ve unfortunately seen is that ICP is considered a one-off exercise versus the continuous development it should be.\n- The Claude bill said $1,492. I paid $20. Here's how. (2026-07-30) https://www.varunchoraria.com/how-to-get-1492-out-of-a-20-claude-subscription/: I pay $20 a month for Claude. Last week I went digging through my own logs to find out what that actually buys, because I wanted to write an honest post about it and I did not want to guess.\n- Hunting Season for the Rest of Us (2026-07-24) https://www.varunchoraria.com/hunting-season-for-the-rest-of-us/: About a week into prepping for one of these companies, I could drop the category’s favourite piece of jargon into a sentence without flinching. Felt good about it too.\n- I now have my own Master Shifu (2026-07-02) https://www.varunchoraria.com/i-now-have-my-own-master-shifu/: I’ve been collecting consulting casebooks for about a decade. Kellogg, Wharton, Ross, Darden, IIM (A, B, C, I, L), MDI, Drucker, McKinsey, Victor Cheng, Crack the Case — 19 books in total, some of them 300+ pages.\n- Maybe going around CIRCLES is worth it (2026-06-26) https://www.varunchoraria.com/maybe-going-around-circles-is-worth-it/: Frameworks get a bad rep. But early in your problem-solving journey, they’re scaffolding — not a crutch. Over time, as you get acquainted with more real-life scenarios, your intuition gets powerful enough to spot-check and work bottoms-up."
    },
    {
      "type": "page",
      "title": "growth",
      "slug": "tags/growth",
      "url": "https://www.varunchoraria.com/tags/growth/",
      "intro": "I write about sustainable growth, go-to-market systems, pipeline, experimentation, and revenue.",
      "listing": true,
      "content": "I write about sustainable growth, go-to-market systems, pipeline, experimentation, and revenue.\n- Maybe going around CIRCLES is worth it (2026-06-26) https://www.varunchoraria.com/maybe-going-around-circles-is-worth-it/: Frameworks get a bad rep. But early in your problem-solving journey, they’re scaffolding — not a crutch. Over time, as you get acquainted with more real-life scenarios, your intuition gets powerful enough to spot-check and work bottoms-up."
    },
    {
      "type": "page",
      "title": "RSS Feed — Subscribe to Varun Choraria's Notes",
      "slug": "feed",
      "url": "https://www.varunchoraria.com/feed/",
      "intro": "Subscribe to every new essay in an RSS reader. No algorithm, ads, email signup, or tracking.",
      "listing": false,
      "content": "Subscribe to get new notes delivered straight to your RSS reader. No algorithms deciding what you see, no ads, no tracking. Just the writing.\n\nFeed URL\n\nCopy this into your RSS reader of choice:\nhttps://www.varunchoraria.com/feed.xml\n\nHow to subscribe\n\n\n  Copy the feed URL above\n  Open your RSS reader\n  Find \"Add feed\" or \"Subscribe\" and paste the URL\n\n\nRSS readers worth trying\n\nI'm a fan of software that's open-source and free — NetNewsWire is what I personally use. It's native Mac and iOS, completely free, and open-source. If you're on Windows or want a web-based option, there are solid alternatives below.\n\n\n  NetNewsWire (Mac, iOS) — Free and open-source. My personal pick.\n  NewsBlur (web, iOS, Android) — Open-source and self-hostable\n  Inoreader (web, iOS, Android) — Powerful filters and rules\n  Feedly (web, iOS, Android) — Good free tier, widely used\n\n\nFrequently asked questions\n\nWhat is RSS?\nRSS (Really Simple Syndication) is an open web standard that lets you follow websites without signing up, giving away your email, or trusting an algorithm to decide what you see. Your reader checks every feed you subscribe to and shows you everything published, in order. No curation, no account required, no inbox to clutter.\n\nIs RSS still relevant?\nYes — arguably more so now. When social feeds are increasingly shaped by engagement algorithms and email inboxes are noisy, RSS is the cleanest way to read what you actually chose to follow. It's how I keep up with writers and blogs I care about without depending on a platform to surface them.\n\nWhat do you write about?\nB2B SaaS marketing, go-to-market strategy, product marketing, sales enablement, AI in B2B revenue teams, management, and career. Mostly longer essays — things I'd want to have read earlier in my career.\n\nHow often do you publish?\nIrregularly. When something is worth writing, not on a schedule. RSS is actually the best way to follow along for that reason — you won't miss a post and you won't get notified when there's nothing new."
    },
    {
      "type": "page",
      "title": "Connect via MCP",
      "slug": "mcp",
      "url": "https://www.varunchoraria.com/mcp/",
      "intro": "Connect an AI client to this site's live, searchable writing and public data through MCP.",
      "listing": false,
      "content": "Let your AI read this site — posts, pages, everything — live, not from training data.\n\n\n  https://varunchoraria-mcpvercelapp.vercel.app\n\n\nWhat to ask once connected\n\n\n  \"What's Varun's take on AI-first GTM strategy?\"\n  \"Summarise his notes on management and team leadership.\"\n  \"What tools and software does he use day to day?\"\n  \"What's his most recent post about?\"\n  \"What side projects has he shipped with AI?\"\n\n\nSee it in action\n\nReal Claude Code sessions using the MCP server — no copy-pasting, no manual context.\n\n\n  \n    \n  \n  \"Use varunchoraria MCP and tell me the top 3 insights across his career blog posts.\" Claude fetched the posts directly and surfaced three sharp takeaways — including \"Luck = Do × Tell (50:50)\" from I've been trying to get lucky for a decade.\n\n\n\n  \n    \n  \n  \"Where does he work, and summarise his career trajectory for me.\" Claude pulled the Work page and mapped the full arc: student founder → founding PMM → PLG operator → category-creating marketing manager.\n\n\nHow this is different from Ask Claude / Ask ChatGPT\n\nThe \"Ask Claude\" and \"Ask ChatGPT\" buttons on this site open a web chatbot with a pre-typed query. That's useful for a quick introduction — but those AIs draw on training data, which has a cutoff date and doesn't include recent posts or pages added after their last training run.\n\nMCP is for a different use case. When you wire up an AI agent in Cursor or Claude Code, it gets live read access to everything on this site — content published today included. Instead of the AI guessing at what I've written, it reads the actual text directly. The result is more accurate, more recent, and fully cited.\n\nUse the chatbot buttons for discovery. Use MCP when you're doing real work — building something, writing a brief, doing research — and want my writing as active context in your workflow, not a vague memory from training.\n\nWhy this exists\n\nPersonal websites are dark matter to AI. Most models have a knowledge cutoff, don't index small personal sites deeply, and go stale fast. Ask a chatbot what I've written about B2B go-to-market strategy and it'll give you a generic answer — not the specific frameworks and cases I've actually published.\n\nMCP is the protocol that fixes this. By running a small server that speaks MCP, any compatible AI agent can access what's here in real time. No copy-pasting, no RAG pipeline, no stale embeddings. The content updates automatically whenever I push new writing — the server always reads from the live site.\n\nThis is also an experiment. I'm interested in what it looks like when a personal site becomes a first-class data source for AI agents, not just a URL to paste into a chat window.\n\nHow it stays current\n\nThere is no crawler and no copy of this site sitting in a database somewhere. Every publish rebuilds a single feed at /api/site.json holding the full text of every post and page, and the server reads that feed live, refreshing at least every five minutes. Publish something now and an agent can read it within the same coffee break.\n\nThe failure mode I actually worried about was quieter: a new page that forgets to opt in, still visible to humans, invisible to agents. So the build compares the sitemap against the feed and fails the deploy if any public URL is missing or exposes no content. If you can reach a page in a browser, an AI client can read it.\n\nIf the site is ever down, the server keeps serving its last good copy rather than erroring. Check what it currently sees, including when the site was last built.\n\nWhat the AI can access\n\n\n  \n    get_site_info\n    Bio, role, topics, contact links\n  \n  \n    list_posts\n    All notes — titles, dates, tags\n  \n  \n    get_post\n    Full text of any post by slug\n  \n  \n    list_pages\n    About, Work, Side Quests, tags, and more\n  \n  \n    get_page\n    Full content of any site page\n  \n  \n    search\n    Ranked full-text search with snippets"
    },
    {
      "type": "page",
      "title": "Work",
      "slug": "work",
      "url": "https://www.varunchoraria.com/work/",
      "intro": "Roles, responsibilities, and selected consulting work.",
      "listing": false,
      "content": "This page is the full record: roles, scope, teams, and outcomes. I’m open to senior full-time roles anywhere in the world, and LinkedIn is the best way to start that conversation. If you’d rather hire the work than the person, here’s how I run engagements.\n\nOutcomes are stated as rates, multiples, and adoption rather than as a former employer’s revenue. The absolute figures are on my resume, which I’m happy to share.\n\nIf you want to know how I think before we talk, start with the essays or what I build on the side.\n\n\n\nCurrent\n\nB2B SaaS Marketing Consultant\n\n2026-present\n\n\n  Advising B2B SaaS companies on growth strategy, pipeline, positioning, product marketing, and AI\n  Running workshops, defined projects, and ongoing marketing leadership\n  Building AI-first marketing systems with small teams and agents\n\n\n\n\nPast\n\nSenior Manager, Marketing\n\nJan 2026 - Aug 2026 · GTM Buddy · Bangalore\n\n\n  Led an inherited team of 3 (Performance Marketer, Designer, SalesOps) and 4 agencies that sourced $2M+ in pipeline on a $500K annual marketing budget\n  Transformed the team to AI-first operations, integrating GitHub, Cursor, Claude, Codex, and Swan into daily work: automated sales signal detection, account research, and meeting prep\n  Designed and built a company-wide GTM data dashboard unifying pipeline, campaign performance, and revenue attribution, which became the single source of truth for investors and leadership\n  Launched Nucleus end-to-end: positioning, messaging, sales enablement, and go-to-market strategy, securing 20 customers in alpha\n  Redesigned the website and brand identity, which became the #1 lead generation channel with a 65% increase in branded search traffic\n  Built automated GTM workflows via Swan: website deanonymization into automated AE first-touch, pipeline health monitoring on Slack, and account research and meeting-prep automation\n  Created and filed a trademark for Revenue Activation, a new category at the intersection of revenue operations and customer activation\n  Founded the SDR motion from scratch, generating 2 qualified meetings in the first month\n  Managed founder brand: 35+ podcast appearances in one quarter (800K+ combined views), LinkedIn grown to 200K+ impressions, and a Forbes feature placement\n  Drove 9 qualified leads through AEO and GEO, positioning the company inside AI-generated answers across ChatGPT, Perplexity, and Claude\n\n\nLead, Product Marketing (PLG)\n\nMay 2024 - Jan 2026 · Freshworks Inc (FRSH) · Bangalore\n\n\n  Drove expansion ARR through PLG campaigns with $0 in paid media\n  Ran the PLG campaign across Top, Mid, and Digital tiers for CX products: 2,778 PQLs converting at 17.5% into paying customers\n  Led the competitive campaign against Intercom for Freshdesk Omni, converting at 11.6%, which was 5.7x the average\n  Ran the multi-competitor “Switch to Freshsales” campaign, generating 775 qualified leads across 22,382 sessions\n  Set up the PLG motion for Freshsales, and developed the market launch strategy for Freddy AI Agent\n  Championed adoption of ChatGPT and built Freddy Sidekick, a custom GPT for 40+ PMMs trained on internal knowledge: >95% team adoption within weeks, cutting research time by 60-70%\n  Helped build the CX PLG charter for 2025\n  Spearheaded the demo experience transformation, prototyping the resource hub and demo centre to shorten the self-serve buyer journey\n  Won the Mid Tier Connect 2025 Award for contributions to PLG\n\n\nProduct Marketing & GTM, Americas\n\nOct 2022 - Apr 2024 · Vymo Inc · Chicago, IL\n\n\n  Relocated to Chicago as the sole product marketer for North America\n  Owned sales enablement that influenced enterprise ARR inside a single quarter\n  Led solution packaging and product launches with solution engineers, sales, and product, driving 85% adoption across 400K+ end users in 7 countries\n  Defined ICP and GTM verticalisation, helping Vymo launch into multiple lines of business across 75+ enterprise deployments\n  Led competitive intelligence: Vymo named a Strong Performer in the Q3 2022 Forrester Wave, and featured in the Gartner Market Guide for Sales Enablement 2023\n  Won the FY23 Annual Award for product marketing work behind the first multi-million dollar deal win in the USA\n\n\nAssociate, Product Marketing & GTM, India & SEA\n\nApr 2020 - Aug 2022 · Vymo Technologies India · Bangalore\n\n\n  Cofounded the product marketing and GTM function\n  Led the launch of the sales enablement and content repository solution: 100% adoption within 15 days, and a 2x improvement in sales collateral adoption\n  Built a no-code cross-platform app for lead capture at event booths across 3 geographies, cutting time-to-first-meet for AEs to under 24 hours\n  Led packaging, positioning, and messaging for all key banking, insurance, and non-banking financial solutions\n  Won a Peer Award for exceptional contributions, and was promoted with an Outstanding Rating by the VP of Product Marketing\n\n\nGrowth Marketer, Content Marketing & Customer NPS\n\nJun 2019 - Mar 2020 · Vymo Technologies India · Bangalore\n\n\n  Led customer NPS campaigns across Capterra and G2, taking Vymo to 4.7/5 in 6 months, an 89% increase\n  Ghostwrote 3 thought leadership articles for the co-founder and CTO, published on the Forbes Technology Council\n  Set up the content writing function, hiring and managing a team of 3 remote freelance writers\n\n\nCo-Founder & CEO\n\n2016 - 2018 · WiseWords Interactive Studio · Bangalore · archived site\n\n\n  Founded the agency at 19, in my first year of university, and grew it to 10+ enterprise and SMB clients\n  Marquee clients included Pall Corp (acquired by Danaher Corp) and Time N Cash Management Consulting\n\n\n\n\nIndependent consulting\n\n\n  Vendasta (Canada, Series D): built an internal sales content repository on Notion, organised by product suite with a context note on every asset, used by 200+ sales agents with 100% adoption\n  Pazo (India, Series A): built the outbound motion from zero for enterprise facilities buyers, from list to sequences, directly impacting 2 enterprise logo wins\n  GTM Buddy (India, Series A): audited every sales asset by stage against rep engagement data, then rebuilt the set: first-call decks, call scripts, onboarding, battlecards, case studies, campaign assets\n  LucioAI (ZenLegal): rebuilt the target definition for a legal AI product after cold email failed against large firms, then ran outbound for events, founder chats and demo signups\n\n\nThe numbers behind these, on request.\n\n\n\nSpeaking, podcasts & publications\n\n\n  Hosted The Edge Podcast for CHROs, interviewing 29 senior leaders from Infosys, Oracle, GoJek, Zoho and others, and co-authored 5 articles on moneycontrol.com about the future of work\n  Delivered sessions on product marketing and B2B GTM at Wynter Games, AIC Prestige, NITI Aayog, AIM, Microsoft for Startups, and Headstart Network\n  Wrote and self-published Miracle-preneuring: Why Anyone Can and Should Be an Entrepreneur\n\n\nMore on my speaking page. The tools I build in public are side quests."
    },
    {
      "type": "page",
      "title": "Tags",
      "slug": "tags",
      "url": "https://www.varunchoraria.com/tags/",
      "intro": "Browse posts by topic.",
      "listing": false,
      "content": "ai\n      11\n    \n  \n    \n    \n    \n    \n    \n      analytics\n      3\n    \n  \n    \n    \n    \n    \n    \n      career\n      8\n    \n  \n    \n    \n    \n    \n    \n      communication\n      1\n    \n  \n    \n    \n    \n    \n    \n      design\n      2\n    \n  \n    \n    \n    \n    \n    \n      frameworks\n      6\n    \n  \n    \n    \n    \n    \n    \n      growth\n      1\n    \n  \n    \n    \n    \n    \n    \n      leadership\n      8\n    \n  \n    \n    \n    \n    \n    \n      management\n      7\n    \n  \n    \n    \n    \n    \n    \n      product\n      1\n    \n  \n    \n    \n    \n    \n    \n      product-marketing\n      3\n    \n  \n    \n    \n    \n    \n    \n      sales\n      2\n    \n  \n    \n    \n    \n    \n    \n      strategy\n      9"
    },
    {
      "type": "page",
      "title": "The first 30 days",
      "slug": "consulting/first-30-days",
      "url": "https://www.varunchoraria.com/consulting/first-30-days/",
      "intro": "I have never opened an engagement with a strategy session. Every one started by finding the belief the team was already paying for that turned out to be wrong.",
      "listing": false,
      "content": "I have never opened an engagement with a strategy session.\n\nFour engagements. GTM Buddy, Vendasta, Pazo, LucioAI. Every one started the same way: find the belief the team was already paying for that turned out to be wrong.\n\nThey split into two shapes.\n\nShape one: the assets exist and nobody uses them\n\nGTM Buddy sold a sales enablement product. Their own reps were winging first calls.\n\nI audited every asset by sales stage against two things they already owned: rep engagement data, and which assets reps actually sent customers through their own digital sales room.\n\nNobody had read that data as a messaging problem. They had read it as an adoption problem.\n\nWhat it showed:\n\n\n  Assets nobody opened.\n  Messaging that changed depending on who wrote it.\n  Deals stalling mid funnel, at exactly the stages where the unused assets sat.\n\n\nOne cause under all three. The messaging spoke to product marketers, sales heads and CROs at once, so it landed with none of them. And it was feature heavy, in a category where nobody buys the feature.\n\nThey needed a value first narrative carrying causal metrics:\n\n\n  Win rate\n  ACV increase\n  Pipeline influenced\n  Ramp time reduction\n\n\nNot adoption of their own product. Adoption is not a result. It is what happens before the result.\n\nThen the rebuild, stage by stage: first call decks, call scripts, an onboarding guide, videos, case studies, battlecards, campaign assets, blogs. QBR decks came later, built with customer success.\n\nVendasta had the same problem inverted. The assets existed and were scattered across everywhere. I built a proxy enablement tool in Notion: one place, organised by product suite, every asset carrying a note on when to use it. Design, implementation, maintenance.\n\nNeither needed a new strategy. Both needed someone to read what was already sitting there.\n\nShape two: nothing to audit, so the target is the risk\n\nPazo and LucioAI had no outbound motion at all. No assets, no data, nothing to read.\n\nSo month one goes into the target definition, because that is the assumption nobody has tested and everybody is about to spend money on.\n\nLucioAI sold to law firms. Law firms do not answer cold email.\n\nThat is not a copy problem. Rewriting the email a fourth time does not fix it.\n\nSo we changed who we scraped:\n\n\n  Small firms, not large ones.\n  UK only, where response rates held up.\n  A sharper promise as the target narrowed: do more with less, at a fraction of what the market leader charges.\n\n\nThen the boring part. Batch the sequences. Route every lead into the CRM. Tell the founders exactly where a human needs to step in.\n\nPazo sold into enterprise facilities management. Same order, different market.\n\nTwo logos won from cold outbound, in a company that had never run it.\n\nWhat month one is actually for\n\nYou are not paying me for an opinion in week one. You are paying me to find out which of your current beliefs is expensive.\n\n\n  If you have assets, I read the engagement data before I read the assets.\n  If you have no assets, I test the target before anyone writes copy.\n  If you have both, I start with whichever one you are most certain about.\n\n\nThat last one is deliberate. The belief nobody questions is the one that has been wrong longest.\n\nWhat you have by day 30\n\n\n  The audit, in writing: which assets are dead, which stages leak, and what each is costing you.\n  First assets shipped, not proposed.\n  A target definition either confirmed against response data or replaced.\n  One number we both agree the next 60 days move.\n\n\nAnd if the answer is that your problem is not marketing, I tell you in week two. That has happened twice out of four.\n\nBook 30 minutes. Bring what you already have, exactly as messy as it is. The mess is the diagnosis."
    },
    {
      "type": "page",
      "title": "Consulting",
      "slug": "consulting",
      "url": "https://www.varunchoraria.com/consulting/",
      "intro": "I help B2B SaaS companies decide what marketing should do next, ship the work, and measure the result in pipeline, revenue, or another business outcome.",
      "listing": false,
      "content": "One person running a swarm of agents. That is the whole engagement.\n\nYou get senior judgment on the decision and the shipped work that follows from it. No account manager, no agency layer, no handoff between the person who made the call and the person who has to execute it.\n\nI work across growth strategy, product marketing, and AI. You can hire me for one hard decision, a defined project, or an ongoing leadership role.\n\nBook 30 minutes on my calendar. Bring the target, problem, or plan you are working through.\n\nHow I actually work\n\nMost consultants sell strategy and hand execution to someone else. Most agencies sell execution and skip the strategy. I do both, because AI collapsed the distance between them.\n\nI run my own marketing this way. I rebuilt this site with agents, turned 19 consulting casebooks into agent commands, and built a research system that does a week of company analysis in an afternoon. I also publish what it costs me to run, down to the token. The tools are open source, so you can inspect the workmanship before you hire it.\n\nThat matters for two reasons. Work that used to need a team now ships from one person at senior quality. And when I recommend AI inside your marketing, I am describing how I already operate, not reading from a deck.\n\nWhat you can hire me to do\n\nPosition the product and make it easier to sell\n\nChoose the ICP, problem, category, value, proof, and message order. Turn internal product knowledge into a story buyers understand and sales can repeat.\n\nThis covers positioning, messaging, launches, sales enablement, competitive work, PLG, and lifecycle. The output is not one messaging document. It is a clearer product story that shows up in the website, launch, campaign, sales call, product, and pipeline.\n\nBuild a growth and pipeline plan\n\nWork backward from the revenue goal. Set a believable pipeline target. Choose the channels and campaigns worth funding. Build the funnel view, budget, forecast, and review process needed to run the plan.\n\nYou get a clear answer to three questions: what are we doing, why should it work, and what will tell us if it is working?\n\nMake AI useful inside marketing\n\nFind where AI can improve the work, build the system around it, train the people who will use it, and measure whether it made the output or the business result better.\n\nThat means AI strategy and team design, agents and workflows connected to real work, AEO and GEO, research and reporting systems, and the review steps that keep quality honest. The point is not more tools. It is more useful work from the team you already have.\n\nLead marketing without a full-time hire\n\nSet priorities, run weekly reviews, manage the team and agencies, watch pipeline, help with hiring, and give founders or leadership one senior person accountable for the whole plan.\n\nTwo things that go wrong most often\n\nAI does not fix unclear marketing\n\nGive a strong model a weak brief and you get weak work faster.\n\nMost failed AI projects are not model problems. Nobody chose the business goal. Context is scattered. The old process was never clear. Review belongs to whoever has time. The team cannot tell whether the new system helped.\n\nBuying tools is easy. Changing how the work gets decided, built, checked, and measured is the job.\n\nThe request is often not the real problem\n\nThe homepage rewrite may be an ICP decision nobody wants to make. The launch plan may be weak because every feature is treated as important. Sales may ignore the battlecard because it does not help with a live deal.\n\nA messaging framework nobody uses is not product marketing. It is a PDF.\n\nWhy teams call me\n\nThe pipeline target and the marketing plan do not add up\n\nThe target came from the board plan. The channel plan came later. Sales, marketing, and finance now use different assumptions about deal size, conversion, sales cycle, and who creates pipeline.\n\nA target is useful. A target built on hidden assumptions is wishful thinking in a spreadsheet.\n\nThe product changed, but the story did not\n\nNew products, segments, and use cases were added. The homepage still tells the old story. Sales has adapted through private decks and calls. Marketing is trying to support every version at once.\n\nThis looks like a copy problem. It is usually a choice problem first.\n\nThe team is busy, but the work does not build on itself\n\nCampaigns ship. Content ships. Reports ship. The next quarter starts from zero because decisions, lessons, and customer evidence live in different heads and documents.\n\nMore output will not fix that. Clear priorities, shared context, and regular review can.\n\nThe company needs senior marketing judgment, but not another full-time leader\n\nThe founder or CMO is the bottleneck. A small team and several agencies need direction. Important work crosses growth, product marketing, sales, and AI.\n\nThe gap is not another pair of hands. It is one person who can make tradeoffs, connect the work, and keep the result visible.\n\nChoose how we work\n\nWorkshop\n\nUse a workshop when the team needs a shared decision before it needs more work. Good topics include AI strategy, positioning, a pipeline plan, a launch, or team design. Practical tool training can be part of it.\n\nThe output is a decision, plan, owners, measures, and next steps. Not inspiration that dies on Monday.\n\nDefined project\n\nUse a project when the outcome is clear enough to name. I lead the thinking and ship the work needed to reach it.\n\nExamples include a positioning reset, product launch, sales enablement system, AI workflow, AEO and GEO program, pipeline plan, or marketing review.\n\nOngoing consulting or fractional leadership\n\nUse an ongoing contract when the work needs senior attention over several cycles. I can set the pace, review pipeline, guide the team, manage key work, train people on the systems, and help the plan improve as the market responds.\n\nThis is not a fallback offer. Some marketing problems only become visible after the first plan meets real buyers.\n\nTalk through the right way to work together.\n\nPipeline numbers belong in the room\n\nA marketing leader who will not discuss pipeline is avoiding the job. A marketing leader who claims full control over revenue is ignoring how B2B SaaS works.\n\nI work between those two positions.\n\nWe agree on the revenue goal, pipeline target, assumptions, attribution rules, and measures before the work starts. Then we report what marketing created or influenced, what converted, what stalled, what the work cost, and what changes next.\n\nAttribution will not be perfect. The number will still be visible. Misses will be explained, not buried inside a slide.\n\nHow the work happens\n\nFind the truth. Review pipeline, customer and sales calls, campaign results, product use, budgets, old plans, and current work.\n\nMake the choices. Agree on the audience, problem, message, channel, target, owner, budget, and what waits.\n\nShip the work. Build the campaign, page, launch, sales material, workflow, report, or plan needed to move the result.\n\nReview and improve. Look at what happened, challenge the assumptions, and change the plan. Ongoing work repeats this loop.\n\nAI runs through all four steps. It does research, call analysis, first drafts, testing, and keeps shared context current. It does not invent customer proof or make the final decision.\n\nWhat we measure\n\nThe right number depends on the work. We choose it before building, because otherwise every project looks successful.\n\n\n  Positioning and website work: message testing, conversion, qualified pipeline, and sales feedback.\n  Launches: reach, interest, adoption, pipeline, revenue, and what the market taught us.\n  Sales enablement: use by reps, deal conversion, sales cycle, objections, and win or loss reasons.\n  PLG and lifecycle: activation, adoption, expansion, retention, and revenue.\n  AI systems: time to publish, review time, quality, cost, use by the team, and the business result linked to the work.\n\n\nProof\n\n\n  At GTM Buddy, I led a team of 3 and 4 agencies that sourced $2M+ in pipeline on a $500K annual marketing budget. I also built the pipeline view used by leadership and investors, and led positioning, messaging, sales enablement, and the launch of Nucleus, which reached 20 alpha customers.\n  At Freshworks, I helped build the PLG charter and ran expansion campaigns that drove ARR with $0 in paid media. The tiered PLG campaign converted 2,778 product-qualified leads at 17.5%, and the competitive campaign against Intercom converted at 5.7x the average.\n  At Vymo, I helped build the product marketing function across APAC, then moved to Chicago as the sole PMM for North America. Over those 5 years, Vymo grew ARR 4x. My work covered product launches, market entry, 85% adoption across 400K+ end users in 7 countries, and the Strong Performer placement in the Forrester Wave.\n  In consulting work, I built a sales content system at Vendasta used by 200+ reps with 100% adoption, rebuilt the sales asset set at GTM Buddy against their own rep engagement data, and ran positioning and outbound at Pazo and LucioAI, tied to 2 enterprise logo wins. What the first 30 days looks like.\n  On AI specifically, I used GitHub, Cursor, Claude, Codex, and Swan in daily marketing work at GTM Buddy, and built systems for sales signals, account research, meeting preparation, campaigns, and pipeline reporting.\n\n\nThe fuller record is on my work page. The systems and experiments I build in public are on GitHub, so you can inspect the work instead of trusting an “AI expert” label.\n\nCommon questions\n\nWho does the work?\n\nI do. That is the point. One senior person working with agents, so the person who makes the decision is the person who ships and reviews the result. If your team or agencies are already in place, I can lead their work or fill a specific gap.\n\nAre you a strategist or an executor?\n\nBoth. I help make the decision and stay close enough to the work to learn whether it was right.\n\nWill you own a pipeline number?\n\nYes, within a clear growth model. We define what marketing owns, what it influences, which assumptions sit with sales or product, and how progress is reported.\n\nCan you run an AI workshop with tool training?\n\nYes. The tools get taught through your work and goals, not as a product tour.\n\nDo you build the AI systems yourself?\n\nYes. I build them, run them on real work, and teach your team how to use and change them.\n\nDo you work with early-stage companies?\n\nYes. The goal changes with the stage. Early work may focus on buyer research, positioning, the offer, and the first repeatable tests. Later work can carry a larger pipeline target, team, and budget.\n\nBring me the real problem\n\nYou do not need to diagnose it before the call. Bring the pipeline model, positioning problem, launch plan, AI question, team problem, or list of competing priorities.\n\nWe will work out whether it needs a workshop, a defined project, or an ongoing role.\n\nBook a 30-minute conversation."
    },
    {
      "type": "page",
      "title": "Speaking & Media",
      "slug": "speaking",
      "url": "https://www.varunchoraria.com/speaking/",
      "intro": "Keynotes, panels, a book, and two seasons of a podcast. The work that happened outside a job description.",
      "listing": false,
      "content": "Book\n\nMiracle-preneuring: Why anyone can and should be an entrepreneur\n\n2016\n\nAvailable on Amazon, Apple Books, and eBay.\n\nWrote, edited, and self-published across platforms. Worked with PR for distribution across 50+ outlets including Forbes, Entrepreneur, and Morning Star. Officially launched at Cafe Coffee Day HQ in Bangalore.\n\nYouTube promos:\n\n\n  #1\n  #2\n  #3\n\n\n\n\nSpeaking\n\nProduct Marketing\n\n\n  (Wynter) Building a Cohesive Powerhouse for B2B Enterprise SaaS\n  GTM-101 For B2B Startups — AIC Prestige, AIM, Niti Aayog, Microsoft For Startups\n  Growth hacking product marketing\n  Product Marketing Alliance: “What Good Looks Like: The Research Inputs Behind High-Performing PLG Funnels” at the Customer & Market Research Summit\n\n\nCareer\n\n\n  Artha Series E3 — Career Frameworks at Jai Hind College, Mumbai\n  Keynote at BIT\n\n\nJudging\n\n\n  Startup event at BIT\n\n\n\n\nPodcast\n\nThe Edge Podcast for CHROs\n\n2021 — co-hosted two seasons exploring the future of work, with guests from GoJek, Infosys, Oracle, MetLife, Ather Energy, Razorpay, Zoho, Doist, and others.\n\nStart here. Eight episodes worth your time, by the company the guest came from:\n\n\n  Oracle: Masterclass on people, process, and productivity, with Shaakun Khanna\n  Infosys: Reskilling and the skills passport, with Richard Lobo, EVP & Head of Human Resources\n  Razorpay: Payroll as a competitive advantage, with Rohit Venugopal of RazorpayX Payroll\n  MetLife: A compensation and benefits playbook, with Srinivas Reddy, Director of HR\n  Ather Energy: Building culture from ground zero, with Sunitha Lal, CHRO\n  GoJek: Marketing to developers, with Sidu Ponnappa, SVP Engineering\n  Doist: Rethinking organisational culture, remote or not, with Andrew Gobran, People Ops\n  Zoho: Welcoming employees back to work, with Praval Singh, VP Marketing\n\n\nSeason 1 focused on return to work and remote work — full playlist\n\nSeason 2 focused on deep dives into specific topics like compensation — full playlist\n\nThree formats: seasonal themes, quick takes (playlist), and Seeing Beyond — futuristic trends (playlist).\n\nMoneycontrol\n\n\n  A guide to how organisations need to reset people practices for “new normal”\n  Future of work: 6 things everyone needs to know about product management\n  5 signs that the creator economy has arrived in India\n  When you go fully remote, terms and conditions apply\n  The varied shades of remote work"
    },
    {
      "type": "page",
      "title": "Michealangelo",
      "slug": "side-quests/michealangelo",
      "url": "https://www.varunchoraria.com/side-quests/michealangelo/",
      "intro": "Ship interfaces that stop looking AI-generated. Give your coding agent design constraints, precise vocabulary, and a taste gate before release.",
      "listing": false,
      "content": "Michealangelo is a free pack of 14 design skills for AI coding agents. It gives Claude Code, Codex, and compatible agents explicit rules for layout, colour, typography, psychology, motion, components, and taste, so interface decisions come from a system instead of the average SaaS page in the model’s training data.\n\nAsk an unconstrained agent for a landing page and familiar defaults appear: centred hero, three cards, gradient button, decorative motion. Michealangelo replaces that pattern matching with constraints you can inspect and reuse.\n\nWhat improves when I use Michealangelo?\n\n\n  Output stops looking interchangeable. taste-audit identifies generic AI interface patterns before they ship.\n  Design decisions come with reasons. Grid, palette, typography, motion, and component choices point back to the rule they followed.\n  Fewer feedback rounds. Precise constraints reduce repeated “not like that” corrections.\n  Sessions stay consistent. design-spec maintains living design tokens and a versioned plan instead of leaving decisions inside an old chat.\n  Updates respect local edits. Hash checks report changed skill files rather than silently overwriting your work.\n\n\nThe benefit is not a prettier first draft. It is a design process that remains arguable, repeatable, and easier to hand off.\n\nHow does Michealangelo work?\n\ndesign-system is the router. Tell it what you are building and it selects the relevant skills in the right order.\n\ngrid-systems sets layout, spacing, and type scale before components appear. color-systems generates palettes and checks contrast. design-psychology covers how people perceive and misread interfaces. component-library checks live registries instead of relying on model memory. motion adds movement only where it clarifies state. taste-audit runs last as the quality gate.\n\nSupporting skills interrogate the brief, pair fonts, sharpen design vocabulary, write handoffs, and preserve design decisions across sessions.\n\nWhy did I build it?\n\nI rebuilt this website with agents. The first attempts looked like every other AI-generated site on the internet.\n\nI do not write code beyond basic HTML, so I could see the problem without being able to repair every detail by hand. What I could do was describe good design precisely enough for a machine to follow: Müller-Brockmann on grids, contrast math instead of colour vibes, and exact language for differences such as kerning versus tracking or affordance versus signifier.\n\nMichealangelo is that judgment turned into files. The wider argument is in How to design without AI slop.\n\nHow do I install Michealangelo?\n\nnpx github:vcxcvii/michealangelo install --target claude\n\n\nChoose codex, agents, or all instead of claude for another target. Install only selected skills with --skills grid-systems,color-systems.\n\n\n  \n    \n      Target\n      Installs to\n      Best for\n    \n  \n  \n    \n      claude\n      ~/.claude/skills/\n      Claude Code\n    \n    \n      codex\n      ~/.codex/skills/\n      Codex\n    \n    \n      agents\n      ./.agents/skills/\n      Project-level compatible agents\n    \n    \n      all\n      All three locations\n      Multiple agents\n    \n  \n\n\nWhat should I not expect?\n\nMichealangelo cannot make you a designer or rescue a brief nobody has thought through. grill-me exists because vague audiences, goals, and constraints create vague output no design system can fully repair.\n\nLatest meaningful changes\n\nThese are product changes, not every repository commit. The list refreshes daily and ignores documentation-only maintenance.\n\n2-jul-2026 · Add design-spec, fix vocab-skill naming · 3 related commits · source for Add design-spec, fix vocab-skill naming\n\nAlso: Add grill-me, michelangelo, and font-pairing skills; Expand into a full self-improving design-skills set\n\n1-jul-2026 · Add document-formatting skill · source for Add document-formatting skill\n\nCaptures ReportLab layout bug fixes (leading defaults, bullet hanging indent, orphaned headers) and an ATS-friendliness checklist learned while reformatting a resume, as a…\n\nQuestions people ask\n\nWhat is Michealangelo?\n\nMichealangelo is a free pack of 14 Agent Skills that gives AI coding agents explicit rules for layout, colour, typography, psychology, motion, components, and design quality.\n\nWhich coding agents support Michealangelo?\n\nIt installs for Claude Code, Codex, and project-level Agent Skills directories used by other compatible coding agents.\n\nDoes Michealangelo generate a design system?\n\nYes. Its router combines relevant skills, while design-spec maintains living design tokens and a versioned specification across sessions.\n\nWill it automatically overwrite my edited skills?\n\nNo. The updater compares file hashes from installation and reports locally edited skills instead of silently replacing them.\n\nFound a design decision that still feels wrong?\n\nOpen an issue on GitHub with the brief, output, and correction you expected. That gap is useful input for the skill set. For a longer conversation, book 30 minutes.\n\nMichealangelo is MIT licensed. Fork it, remove what you do not need, and disagree with it in public."
    },
    {
      "type": "page",
      "title": "Rainmaker",
      "slug": "side-quests/rainmaker",
      "url": "https://www.varunchoraria.com/side-quests/rainmaker/",
      "intro": "A free, open-source SEO and AEO agent that runs inside Codex, Claude Code, and compatible coding assistants. It crawls your site, finds the three fixes closest to revenue, and remembers whether they worked.",
      "listing": false,
      "content": "What is Rainmaker?\n\nRainmaker is a free SEO and AEO agent that works inside your coding assistant. Point it at a website and it will crawl the pages, work out which problems sit closest to revenue, and give you three things worth doing next. It keeps a permanent record, so it can later tell you what improved and what did nothing.\n\nView the source on GitHub or book 30 minutes with me if you want help applying it to a real site.\n\nWhat problem does it solve?\n\nMost SEO audits produce a long list of problems. A missing alt attribute on a careers page can sit beside a noindexed pricing page. Both look urgent. Neither comes with a useful answer to the question that matters:\n\n\n  Which fix is most likely to help this website make money?\n\n\nRainmaker starts there. It sorts every page into a simple revenue tier, computes scores in code, and recommends three fixes instead of sixty. Then it records what was shipped and checks the right metric after enough time has passed.\n\nWhat do I get?\n\n\n  A first audit with no paid account. The built-in crawler works without Firecrawl, Ahrefs, Semrush, or a model API key.\n  Three prioritized fixes. Each recommendation explains the evidence, likely impact, effort, and consequence of doing nothing.\n  A grounded interview. Rainmaker crawls first, then asks one question at a time using facts from your site.\n  SEO and AEO in one workflow. It covers technical problems, search demand, live SERPs, site structure, answer-engine citations, and content quality.\n  Memory across sessions. An append-only ledger records what Rainmaker believed, what changed, what shipped, and what moved.\n  A safer agent. Scores are deterministic. Paid providers require approval. Publishing and other hard-to-reverse actions stay human-controlled.\n\n\nHow does Rainmaker work?\n\n\n  \n    \n      Step\n      What happens\n      What you get\n    \n  \n  \n    \n      1. Crawl\n      The built-in crawler reads the site before asking business questions.\n      A factual diagnosis, even with zero credentials.\n    \n    \n      2. Understand\n      Your assistant asks one evidence-based question at a time about buyers, conversions, competitors, and value.\n      Shared business context instead of generic SEO advice.\n    \n    \n      3. Prioritize\n      Code ranks findings by distance to revenue, opportunity, severity, confidence, and effort.\n      Three fixes worth considering now.\n    \n    \n      4. Implement\n      You choose a fix. The assistant drafts the change or creates the work item.\n      A reviewable diff or issue, not a silent production change.\n    \n    \n      5. Prove\n      Rainmaker checks the relevant metric after its verification window.\n      A record of what worked, regressed, or did nothing.\n    \n  \n\n\nThat final step is the point. A plan becomes stale. A system that changes its mind when the evidence changes becomes more useful.\n\nInstall Rainmaker\n\nInstall the plugin for your assistant (below), open it in your website project, and say:\n\nrun rainmaker\n\n\nThat is the whole setup. The front-door skill asks for your site URL, sets the project up itself, crawls, and opens the conversation from what it found. You never drive the CLI by hand.\n\nIf your assistant has no plugin support, set the project up first and then say the same thing:\n\nnpx @vcxcvii/rainmaker init --site https://yoursite.com\n\n\nRun that in the folder for the site, not your home directory: it writes configuration, context and skill copies into the working directory, and refuses a home directory unless you pass --force.\n\nRainmaker installs one front-door skill plus 26 decision skills. The front door resumes from the first incomplete step. It does not launch a second chatbot or ask for another model key.\n\nInstall the native Codex plugin\n\ncodex plugin marketplace add vcxcvii/rainmaker --ref main\ncodex plugin add rainmaker@vcxcvii\n\n\nStart a new Codex task in your website project and say run rainmaker.\n\nInstall the Claude Code plugin\n\n/plugin marketplace add vcxcvii/rainmaker\n/plugin install rainmaker@vcxcvii\n\n\nClaude Code gets the same front-door workflow plus a session hook that notices unfinished Rainmaker work.\n\nDo I need API keys?\n\nNo key is required for the first useful result.\n\n\n  \n    \n      Credential\n      What it adds\n      Required?\n    \n  \n  \n    \n      None\n      Built-in crawl, technical checks, revenue tiers, scoring, and reports\n      No\n    \n    \n      Google service account\n      Search Console opportunity data and GA4 conversion evidence\n      Optional\n    \n    \n      Clarity\n      Behavioural evidence such as dead clicks and rage clicks\n      Optional\n    \n    \n      Firecrawl or context.dev\n      Alternative crawling for sites that need it\n      Optional and approval-gated\n    \n    \n      OpenAI or Anthropic API key\n      Standalone terminal agent and direct citation probes\n      Optional\n    \n  \n\n\nYour Codex or Claude Code session already supplies the model for the normal workflow. The local CLI cannot borrow subscription credentials from those apps, so it does not try.\n\nWhat can I ask Rainmaker?\n\n\n  “Which three fixes are closest to revenue?”\n  “Our traffic fell. Was it something we shipped, a wider search change, or neither?”\n  “Which pages get impressions but fail to earn clicks?”\n  “Can we realistically rank for this query? Read the live results before answering.”\n  “What did we ship last quarter that produced no measurable improvement?”\n\n\nIt works for a thirty-page personal site, an ecommerce catalogue, a local business, a B2B software company, or a site with thousands of URLs. The revenue model and conversion paths change the advice.\n\nHow is it different from other SEO tools?\n\n\n  \n    \n      Option\n      Good at\n      What Rainmaker adds\n    \n  \n  \n    \n      Ahrefs or Semrush\n      Large search, backlink, and competitor datasets\n      A decision layer that ranks work by business value and remembers outcomes\n    \n    \n      Site audit crawler\n      Finding technical defects\n      Revenue tiers, buyer context, implementation, and verification\n    \n    \n      Folder of SEO skills\n      Giving an assistant specialist prompts\n      One shared strategy, deterministic scoring, and one owner per decision\n    \n    \n      Experienced agency\n      Nuance, stakeholder work, creative judgment, and accountability\n      A transparent local system the agency or in-house team can inspect and reuse\n    \n  \n\n\nRainmaker is strongest at repeatable analysis, prioritization, and memory. It is weaker where the answer lives in customer politics, company context, taste, or judgment that has not yet become evidence. That is deliberate, not hidden.\n\nWhat will Rainmaker never do quietly?\n\n\n  Spend Firecrawl or another paid provider’s quota because a key exists. It asks first, shows you the balance, and remembers your answer.\n  Let a model invent or adjust a revenue score.\n  Publish content, post in communities, send outreach, redirect a page, or delete a URL.\n  Pretend an algorithm update caused a change when the evidence only shows timing.\n  Hide missing Search Console, GA4, or other evidence from the report.\n\n\nLatest meaningful changes\n\nThese are product changes, not every repository commit. The list refreshes daily and ignores documentation-only maintenance.\n\n1-aug-2026 · Rainmaker v0.3.1 · source for Rainmaker v0.3.1\n\nRainmaker now starts and stays inside the assistant you already use. - run rainmaker triggers a dedicated front-door skill - Codex, Claude Code, and compatible hosts conduct…\n\n1-aug-2026 · Rainmaker 0.3.0 · source for Rainmaker 0.3.0\n\nPortable, conversation-first Rainmaker for compatible LLM assistants. Built-in crawling is now the unconditional default; Firecrawl and context.dev require explicit approval…\n\n1-aug-2026 · v0.2.1 · source for v0.2.1\n\nFull Changelog: https://github.com/vcxcvii/rainmaker/commits/v0.2.1\n\nQuestions people ask\n\nWhat is Rainmaker?\n\nRainmaker is a free, open-source SEO and AEO agent for coding assistants. It crawls your website, connects findings to how the business makes money, recommends the three most useful fixes, and records whether those fixes worked.\n\nHow do I start Rainmaker?\n\nInstall the plugin for your assistant, open it in your site project, and say run rainmaker. The skill asks for your site URL and sets everything up itself. If your assistant has no plugin support, run npx @vcxcvii/rainmaker init –site https://yoursite.com in the site project first, then say run rainmaker. Run it in the folder for the site rather than your home directory.\n\nDo I need an OpenAI or Anthropic API key?\n\nNo for the normal workflow. The assistant you already opened conducts the conversation using its existing model session. A separate model API key is only needed for the optional standalone terminal agent and direct AI citation probes.\n\nWhy can Rainmaker not use my ChatGPT or Claude subscription from the standalone CLI?\n\nApp subscriptions do not expose their session credentials or model runtime to child command-line processes. Rainmaker avoids that problem by running the interview inside the assistant itself. The CLI only handles crawling, measurement, scoring, and memory.\n\nDoes Rainmaker use Firecrawl automatically if it finds my key?\n\nNo, and it does not quietly ignore the key either. The built-in crawler is the default. When a Firecrawl or context.dev key is present, Rainmaker reads your live credit balance, tells you what you have, and asks which crawler you want before the first crawl. The answer is saved as crawl.provider in rainmaker.config.yml and honoured afterwards, so you are asked once rather than once per crawl. A crawl projected to exceed your remaining balance is refused.\n\nIs Rainmaker a replacement for Ahrefs, Semrush, or an experienced SEO agency?\n\nIt replaces repetitive crawling, prioritization, and measurement, not every kind of judgment. Ahrefs and Semrush remain useful data sources. An experienced operator is still better at stakeholder alignment, customer nuance, creative positioning, and decisions where the evidence is incomplete.\n\nWill Rainmaker publish content or change my live site by itself?\n\nNo. It can draft, edit local files, and file issues. Publishing, outreach, redirects, deletions, and other externally visible actions require human approval.\n\nWant the system plus human judgment?\n\nRainmaker handles the repeatable work. I can help with the parts that need an experienced operator: positioning, prioritization, buyer nuance, content judgment, and getting a team to act on the evidence.\n\nBook a 30-minute call with me, open an issue, or fork the MIT-licensed project and make it disagree with me in public."
    },
    {
      "type": "page",
      "title": "Lazarus Pit",
      "slug": "side-quests/lazarus-pit",
      "url": "https://www.varunchoraria.com/side-quests/lazarus-pit/",
      "intro": "A self-healing UX agent that reads this site's Clarity sessions every week, diagnoses friction, and files a GitHub issue with a proposed fix. A human still merges.",
      "listing": false,
      "content": "Lazarus Pit is a self-healing UX agent for this website. Every Monday it pulls Microsoft Clarity data, compares the numbers against friction thresholds, and files a GitHub issue naming the problem, the likely cause, and a fix worth trying.\n\nThe repository is private. Its output is not: every issue it has filed is visible in this site’s issue tracker.\n\nWhat problem does it solve?\n\nAnalytics tools are good at telling you something is wrong and bad at making you do anything about it. Clarity records dead clicks, rage clicks, and quickbacks faithfully, then waits for a human to log in, notice, interpret, and act. On a personal site nobody logs in.\n\nLazarus Pit removes the logging-in step. The finding arrives where the work already happens, as an issue on the repository that renders the page.\n\nHow does Lazarus Pit work?\n\nFour modules run in sequence.\n\n\n  Fetch. fetch-clarity.ts calls the Clarity Data Export API for a three-day window and writes the raw response to disk. It tracks its own daily call budget locally and refuses to exceed ten calls a day.\n  Diagnose. finding-extractor.ts scores each metric against a warn threshold and a high threshold, then ranks the findings by severity.\n  Map. component-mapper.ts converts each finding into a titled suggestion: what the metric was, what it probably means, and where to look first.\n  Propose. pr-generator.ts files one GitHub issue per finding on the target repository, labelled lazarus-pit, skipping any finding whose issue is already open.\n\n\nrun.ts chains all four. A GitHub Actions workflow runs the chain every Monday at 14:00 UTC and can also be triggered by hand.\n\nWhat counts as friction?\n\n\n  \n    \n      Metric\n      Warn\n      High\n      Reading\n    \n  \n  \n    \n      Dead clicks\n      5% of sessions\n      15%\n      Something looks clickable and is not\n    \n    \n      Rage clicks\n      3%\n      10%\n      Element is unresponsive or gives no feedback\n    \n    \n      Quickbacks\n      10%\n      25%\n      The link promised something the page did not deliver\n    \n    \n      Script errors\n      1%\n      5%\n      JavaScript failing inside real sessions\n    \n    \n      Error clicks\n      1%\n      5%\n      A click handler is breaking\n    \n    \n      Scroll depth\n      below 60% average\n      below 40%\n      Content past the fold is not being seen\n    \n  \n\n\nWhy does it refuse to diagnose small traffic?\n\nThe first version filed confident findings off eleven sessions, most of them bots. On that volume a single crawler moves a percentage several points and every threshold becomes noise.\n\nSo the agent now subtracts bot sessions from the total and checks the remainder against a floor of twenty human sessions. Below the floor it files one finding and stops: this is a distribution problem, not a friction problem. That single change is the difference between a tool that reports and a tool worth reading.\n\nWhy does it not write the fix itself?\n\nIt could. Filing a patch is not the hard part.\n\nThe hard part is that a site-wide metric cannot tell you which page misbehaved. Clarity’s export API returns numbers for the project, not per URL, so every finding is a site-level hypothesis. Auto-committing against a hypothesis produces confident, wrong diffs that a human then has to reverse. Issues cost a review click and lose nothing.\n\nEscalating to generated pull requests is worth doing once finding quality has earned it. It has not yet.\n\nWhy is the repository private?\n\nNothing in it is secret, but nothing in it is reusable either. Thresholds are tuned to one small site, findings are site-level by necessity, and the fix templates assume this site’s components. Publishing it would ship a tool that mostly teaches people the wrong thresholds for their own traffic.\n\nThe parts worth copying are described on this page. If that changes, so will the repository.\n\nLatest meaningful changes\n\nThese are product changes, not every repository commit. The list refreshes daily and ignores documentation-only maintenance.\n\n1-aug-2026 · Rainmaker v0.3.1 · source for Rainmaker v0.3.1\n\nRainmaker now starts and stays inside the assistant you already use. - run rainmaker triggers a dedicated front-door skill - Codex, Claude Code, and compatible hosts conduct…\n\n1-aug-2026 · Rainmaker 0.3.0 · source for Rainmaker 0.3.0\n\nPortable, conversation-first Rainmaker for compatible LLM assistants. Built-in crawling is now the unconditional default; Firecrawl and context.dev require explicit approval…\n\n1-aug-2026 · v0.2.1 · source for v0.2.1\n\nFull Changelog: https://github.com/vcxcvii/rainmaker/commits/v0.2.1\n\nQuestions people ask\n\nIs Lazarus Pit open source?\n\nNo. The repository is private. Its output is public, because every issue it files lands on this website’s public repository under the lazarus-pit label.\n\nDoes Lazarus Pit change code by itself?\n\nNo. It files a GitHub issue containing the metric, the hypothesis, and a suggested fix. A human reviews, edits, and merges. There is no automatic commit or deploy.\n\nWhat data does Lazarus Pit read?\n\nOnly the Microsoft Clarity Data Export API for this site, which returns site-wide metrics such as dead clicks, rage clicks, quickbacks, scroll depth, and script errors. It does not read session recordings.\n\nWhy does it run weekly instead of daily?\n\nClarity’s export API allows ten calls a day and a three-day lookback. One weekly run stays inside both limits and matches how quickly a small site accumulates enough human sessions to diagnose.\n\nWant to see what it caught?\n\nIts findings are public: browse the open and closed issues it filed against this site. If you are building something similar and want to compare thresholds, book 30 minutes."
    },
    {
      "type": "page",
      "title": "Master Shifu",
      "slug": "side-quests/master-shifu",
      "url": "https://www.varunchoraria.com/side-quests/master-shifu/",
      "intro": "Find the cause before proposing the fix. Give your AI agent 32 consulting frameworks, a question-first workflow, and recommendations you can defend.",
      "listing": false,
      "content": "Master Shifu is a free pack of 32 consulting frameworks for AI agents. Describe a business problem and its router chooses the right frameworks, asks for missing evidence, analyses the problem through five business lenses, and produces a recommendation you can inspect, challenge, and defend.\n\nMost AI answers fail before the writing starts. The agent accepts the first framing, skips root-cause analysis, and produces a polished recommendation for the wrong problem. Master Shifu makes the reasoning happen before the answer.\n\nWhat changes when I use Master Shifu?\n\n\n  You get a cause, not a guess. Profitability, pricing, market entry, M&A, cost reduction, and other frameworks isolate where the problem sits.\n  Missing evidence becomes visible. The question-first grill exposes assumptions, absent trends, and weak data before analysis begins.\n  Recommendations have a structure. Customer, competition, financial, operations, and market lenses make the conclusion easier to audit.\n  You can defend the work in a room. The output shows how the answer was reached, not only what the answer is.\n  Case practice behaves more like an interview. The agent asks, waits, tests, and routes instead of immediately revealing a model answer.\n\n\nBetter structure does not guarantee the right decision. It gives you something concrete to disagree with before the decision becomes expensive.\n\nHow does Master Shifu work?\n\n/master-shifu is the router. You describe the situation in plain language. It identifies useful frameworks, explains the order, and asks for the evidence each framework needs.\n\nThe pack covers 32 recurring business problem shapes extracted from 19 MBA casebooks. Frameworks can run alone or chain together. A margin decline might begin with profitability, move into cost structure, and end with pricing. A market-entry question may need industry analysis before the entry framework is useful.\n\nWhy did I build it?\n\nI spent years being told to be “more strategic” without anyone defining what that meant. The useful answer was unglamorous: strategy is mostly structured decomposition, and structured decomposition can be learned.\n\nI read 19 casebooks and kept the frameworks that repeatedly helped separate symptoms from causes. Turning them into Agent Skills made them available under pressure, when memory usually reaches for the most familiar framework instead of the right one.\n\nThe related essay is “You’re not strategic enough”, said Manager. The build story is in I now have my own Master Shifu.\n\nHow do I install Master Shifu?\n\ncurl -fsSL https://raw.githubusercontent.com/vcxcvii/master-shifu/main/install.sh | bash\n\n\nOr:\n\nnpx skills add vcxcvii/master-shifu\n\n\nIt works with Claude Code, Codex, Cursor, Windsurf, and other tools supporting the Agent Skills format. Then describe the problem: “our SaaS margins fell 12 percent this quarter.”\n\nWhat should I not expect?\n\nMaster Shifu cannot know your business, repair missing data, or rescue a conclusion you already chose and want justified. A framework is a lens, not evidence. You remain responsible for the inputs and the decision.\n\nLatest meaningful changes\n\nThese are product changes, not every repository commit. The list refreshes daily and ignores documentation-only maintenance.\n\n1-jul-2026 · Add stress test suite · 2 related commits · source for Add stress test suite\n\nAlso: Master Shifu - 32 consulting framework skills with OKF format, Scrapling monitoring, task-observer meta-skill\n\nQuestions people ask\n\nWhat is Master Shifu?\n\nMaster Shifu is a free pack of 32 consulting frameworks converted into Agent Skills for structured business analysis and case-interview practice.\n\nWhich business problems can it analyse?\n\nIt covers profitability, pricing, market entry, M&A, cost reduction, industry analysis, growth, operations, and other common strategy problems.\n\nDoes Master Shifu choose the framework for me?\n\nYes. The master-shifu router matches the problem to suitable frameworks, asks for missing evidence, and chains multiple frameworks when needed.\n\nCan it replace a consultant or operator?\n\nNo. It structures evidence and reasoning. It cannot know private context, validate weak inputs, or replace accountable human judgment.\n\nUsed it on a real problem?\n\nOpen an issue on GitHub with the problem shape, chosen framework, and where the reasoning broke. For a longer conversation, book 30 minutes.\n\nMaster Shifu is MIT licensed. Pair it with Interview Recon when a company-specific interview needs both research and case frameworks."
    },
    {
      "type": "page",
      "title": "Interview Recon",
      "slug": "side-quests/interview-recon",
      "url": "https://www.varunchoraria.com/side-quests/interview-recon/",
      "intro": "Walk into interviews with a sourced company dossier, JD-mapped talking points, sharper questions, and a grounded 90-day plan.",
      "listing": false,
      "content": "Interview Recon is a free AI interview research skill for candidates using Claude Code, Codex, Cursor, or another Agent Skills-compatible agent. It turns a job description and public company sources into a reusable dossier: JD-mapped talking points, a grounded 90-day plan, credible questions, and citations you can verify before the call.\n\nMost candidates skim a company website the night before. Real research takes days, so everyone arrives repeating the same surface-level facts. Interview Recon compresses that work into an afternoon while keeping the evidence attached.\n\nWhat will I walk into the interview with?\n\n\n  Talking points tied to the job description. Each requirement connects to company evidence and a point you can defend.\n  A 90-day plan grounded in current moves. Hiring, product, pricing, competitors, funding, and public-market signals shape the plan.\n  Questions other candidates will not ask. The strongest questions come from tensions in the company’s own strategy, not generic interview lists.\n  Claims you can verify. Every factual claim carries a numbered source, URL, and fetch date.\n  Research that survives the next round. The dossier persists, refreshes stale material, and records what earlier interviewers asked.\n\n\nThe output is designed for use in the room, not for producing the longest possible research document.\n\nHow does Interview Recon work?\n\n\n  Save the job description first. Listings disappear. Your copy becomes the spine of the research.\n  Approve a short research plan. The skill names its questions, sources, backend, and budget before fetching.\n  Research the company in a fixed order. It covers products, pricing, competitors, content, leadership, hiring, news, funding, and relevant earnings calls.\n  Synthesize a candidate dossier. You get a one-page brief, strategy canvas, go-to-market teardown, people signals, interview kit, and source log.\n\n\nPublic neighbours matter when the target company is private. Their earnings calls expose market pressures and narratives your interviewer may already be working inside.\n\nWhy did I build it?\n\nI built Interview Recon during my own 2026 job hunt. I was running several interview processes at once, losing the source behind useful facts, and repeating research before every round.\n\nPersistence fixed that. Run the skill again and it loads what it already knows, refreshes only what went stale, and appends what the last round actually asked. The full build story, with screenshots, is in Hunting Season for the Rest of Us.\n\nDo I need a paid scraping API?\n\nNo. Without a key, Interview Recon uses your agent’s built-in web search and fetch. The workflow and outputs stay the same. It flags JavaScript-heavy or bot-protected pages it could not fully inspect instead of pretending the research is complete.\n\nWith a Firecrawl key, it can use faster scraping. It announces a credit budget before starting: 15 credits in quick mode or 40 in deep mode. It never exceeds that budget without asking.\n\nHow do I install Interview Recon?\n\nFor Claude Code:\n\ngit clone https://github.com/vcxcvii/interview-recon.git ~/.claude/skills/interview-recon\n\n\nOpen Claude Code and say, “prep me for my interview at [company].” For another Agent Skills-compatible tool, place the repository in that agent’s skills directory.\n\nWhat should I not expect?\n\nInterview Recon cannot see private company information, replace your judgment, or make you sound like someone you are not. It turns public evidence into better preparation. You still decide what is true, relevant, and worth saying.\n\nLatest meaningful changes\n\nThese are product changes, not every repository commit. The list refreshes daily and ignores documentation-only maintenance.\n\n24-jul-2026 · Interview Recon v1: candidate-POV company research skill for AI agents · source for Interview Recon v1: candidate-POV company research skill for AI agents\n\nQuestions people ask\n\nIs Interview Recon free?\n\nYes. Interview Recon is MIT licensed. It works with your agent’s built-in web tools, so a paid scraping API is optional.\n\nWhich AI agents support Interview Recon?\n\nIt works with Claude Code, Codex, Cursor, and other agents that support the Agent Skills format.\n\nWhat does Interview Recon produce?\n\nIt produces a sourced company dossier, strategy canvas, go-to-market teardown, people signals, JD-mapped talking points, a 90-day plan, and questions to ask.\n\nDoes it reuse research between interview rounds?\n\nYes. It loads the existing dossier, refreshes stale material, and appends notes from each round instead of starting over.\n\nWant to improve the next interview kit?\n\nOpen an issue on GitHub with a missing source, broken workflow, or output that did not help. For a longer conversation, book 30 minutes.\n\nInterview Recon is MIT licensed. Fork it, adapt it, and keep your research. If you also need case-interview frameworks, Master Shifu chains into it."
    },
    {
      "type": "page",
      "title": "Side Quests",
      "slug": "side-quests",
      "url": "https://www.varunchoraria.com/side-quests/",
      "intro": "Projects, tools, and rabbit holes. Most started as weekend curiosity. A few became things I use every day.",
      "listing": false,
      "content": "Personal\n    \n      \n        \n          \n            \n              \n                \n              \n                \n\n\n\n  \n\n\n\n                Interview Recon\n              \n            \n            Public\n          \n          Turns your AI coding agent into an interview research analyst. Candidate-POV company dossiers, JD-mapped talking points, and a 90-day plan. Free-tier friendly by design. Interview Recon source code\n        \n      \n        \n          \n            \n              \n                \n              \n                \n\n\n\n  \n\n\n\n                Master Shifu\n              \n            \n            Public\n          \n          Consulting case framework skills for AI agents. 32 frameworks extracted from 19 MBA casebooks. Master Shifu source code\n        \n      \n        \n          \n            \n              \n                \n              \n                \n\n\n\n  \n\n\n\n                Michealangelo\n              \n            \n            Public\n          \n          Self-improving design-system skills for Claude, Codex, and other Agent Skills-compatible coding agents. Michealangelo source code\n        \n      \n        \n          \n            \n              \n                \n              \n                \n\n\n\n  \n\n\n\n                Rainmaker\n              \n            \n            Public\n          \n          Open-source SEO and AEO agent that ranks every finding by distance to revenue, checks the live SERP before briefing anything, and keeps a permanent record of whether the fix worked. Rainmaker source code\n        \n      \n        \n          \n            \n              \n                \n              \n                \n\n\n\n  \n\n\n\n                VC's Notes\n              \n            \n            Public\n          \n          This lightweight Jekyll publication about AI, GTM, management, careers, and side quests.\n        \n      \n        \n          \n            \n              \n                \n              \n                \n\n\n\n  \n\n\n\n                Self-updating GitHub profile\n              \n            \n            Public\n          \n          Profile README that refreshes itself from current GitHub activity.\n        \n      \n        \n          \n            \n              \n                \n              \n                \n\n\n\n  \n\n\n\n                MCP server\n              \n            \n            Public\n          \n          Lets MCP-compatible AI clients read this site's current posts and pages directly.\n        \n      \n    \n  \n\n\n    Experiments\n    \n      \n        \n          \n            \n              \n                \n              \n                \n\n\n\n  \n\n\n\n                Lazarus Pit\n              \n            \n            Private\n          \n          Self-healing UX agent that reads Clarity sessions, diagnoses friction, and files proposed fixes as GitHub issues on this site.\n        \n      \n    \n  \n\n\n    Work\n    \n      \n        \n          \n            \n              GTM Buddy Marketing Skills\n            \n            Private\n          \n          Shared marketing skills and context for consistent SEO, CRO, copy, ads, analytics, email, and research work.\n        \n      \n        \n          \n            \n              GTM Buddy Design and Engineering\n            \n            Private\n          \n          Reusable agent workspace for building, testing, and shipping responsive GTM Buddy web pages.\n        \n      \n        \n          \n            \n              GTM Skills (SDR)\n            \n            Private\n          \n          Routed SDR skills for cold calls, LinkedIn messages, email sequences, and competitive displacement."
    },
    {
      "type": "page",
      "title": "Blog",
      "slug": "blog",
      "url": "https://www.varunchoraria.com/blog/",
      "intro": "Every essay, newest first, grouped by year.",
      "listing": true,
      "content": "Every essay, newest first, grouped by year.\n- My agent found 9.09%. I found one session. (2026-08-03) https://www.varunchoraria.com/11-sessions-are-not-a-ux-audit/: My agent found a UX problem on my site. Dead clicks had affected 9.09% of sessions, above the 5% threshold. It filed a GitHub issue and told me what to inspect.\n- The models were fine. I was the bug. (2026-08-02) https://www.varunchoraria.com/the-models-were-fine-i-was-the-bug/: I typed a command and nothing happened. No output, no error, no prompt coming back, just a cursor sitting there while my laptop got warm.\n- Who do we sell to vs. Who SHOULD we sell to (and how). (2026-08-01) https://www.varunchoraria.com/who-do-we-sell-to-vs-who-should-we-sell-to-and-how/: I’ve been a part of both startups and large organizations, and one consistent trend I’ve unfortunately seen is that ICP is considered a one-off exercise versus the continuous development it should be.\n- The Claude bill said $1,492. I paid $20. Here's how. (2026-07-30) https://www.varunchoraria.com/how-to-get-1492-out-of-a-20-claude-subscription/: I pay $20 a month for Claude. Last week I went digging through my own logs to find out what that actually buys, because I wanted to write an honest post about it and I did not want to guess.\n- Hunting Season for the Rest of Us (2026-07-24) https://www.varunchoraria.com/hunting-season-for-the-rest-of-us/: About a week into prepping for one of these companies, I could drop the category’s favourite piece of jargon into a sentence without flinching. Felt good about it too.\n- Github-ing (2026-07-16) https://www.varunchoraria.com/github-ing/: The world is moving toward systems-first work, and I do not think most people understand what that means yet.\n- How to design without AI Slop (2026-07-08) https://www.varunchoraria.com/how-to-design-without-ai-slop/: You are what you admire\n- I now have my own Master Shifu (2026-07-02) https://www.varunchoraria.com/i-now-have-my-own-master-shifu/: I’ve been collecting consulting casebooks for about a decade. Kellogg, Wharton, Ross, Darden, IIM (A, B, C, I, L), MDI, Drucker, McKinsey, Victor Cheng, Crack the Case — 19 books in total, some of them 300+ pages.\n- Nobody uses my MCP server. Building it still worked. (2026-06-28) https://www.varunchoraria.com/nobody-uses-my-mcp-server-building-it-still-worked/: Nobody’s going to read this, and I don’t mean this post specifically. I mean the whole thing: the blog, the about page, the work history. A hiring manager has 400 applications open and 11 minutes. A reader lands on one essay from a link ...\n- Maybe going around CIRCLES is worth it (2026-06-26) https://www.varunchoraria.com/maybe-going-around-circles-is-worth-it/: Frameworks get a bad rep. But early in your problem-solving journey, they’re scaffolding — not a crutch. Over time, as you get acquainted with more real-life scenarios, your intuition gets powerful enough to spot-check and work bottoms-up.\n- Too much parenting, too little adulting (2026-06-20) https://www.varunchoraria.com/too-much-parenting-too-little-adulting/: In the age of agentic AI, something I’ve first-hand witnessed as a people manager is that the default mental models are unchanged.\n- Killed By Google (2026-06-16) https://www.varunchoraria.com/killed-by-google/: I first came across killed by google and was honestly quite blown. I suppose the startup myth is real: your instinct is right 95% of the time, but your ideas are wrong probably 80% of the time.\n- Who owns what? (2026-05-17) https://www.varunchoraria.com/who-owns-what/: Everyone these days is asking the same question in the age of agentic AI: what should an AI do vs. what should a human do?\n- Team values I value (2026-04-25) https://www.varunchoraria.com/team-values-i-value/: We work with a bunch of vendors as a company of 44 people. As a manager, I have 2 choices:\n- I've been trying to get lucky for a decade. Here's what I learnt. (2026-04-19) https://www.varunchoraria.com/ive-been-trying-to-get-lucky-for-a-decade/: I think one of the things I have a perpetual obsession with is career optimization. It feels a lot like poker. There is no magic playbook, or a clean meritocratic path to it, at least not anymore.\n- Imposter syndrome as a first-time manager (2026-04-12) https://www.varunchoraria.com/imposter-syndrome-as-a-first-time-manager/: It’s been about a quarter since I became a manager. I’m wondering what value I’m actually adding to the organization, especially in the agentic era where skills, information, and knowledge are more or less commoditized. Here are my candi...\n- So you want to help your team succeed (2026-02-07) https://www.varunchoraria.com/so-you-want-to-help-your-team-succeed/: It’s been about a month since I’ve been a first-time manager. Before I started, I asked my CEO what advice he had for me. Here’s what he said:\n- \"You're not strategic enough\", said Manager (2025-10-26) https://www.varunchoraria.com/youre-not-strategic-enough-said-manager/: Cracks\n- 6 pro tips on communication that's sure to get you promoted (2025-06-15) https://www.varunchoraria.com/6-pro-tips-on-communication-thats-sure-to-get-you-promoted/: I’ve been in the workforce for almost a decade now, and I’ve observed a rather persistent truth: being a great IC gets you noticed, but becoming a great communicator is what gets you promoted.\n- What your career path eventually comes down to (2025-03-08) https://www.varunchoraria.com/what-your-career-path-eventually-comes-down-to/: I’ve spent most of the last 6 years trying to find the answer. Along the way, I’ve seen a pretty clear difference between the low and the top performers.\n- Problems tell you what to fix. Pain points tell you why it matters. (2025-01-30) https://www.varunchoraria.com/problems-tell-you-what-to-fix-pain-points-tell-you-why-it-matters/: Every marketer, founder, or sales professional starts with the same question: “What’s the pain point of my prospect?” But too often, we conflate pain with problems, and that mistake costs us deals, loyalty, and growth."
    },
    {
      "type": "page",
      "title": "Home",
      "slug": "home",
      "url": "https://www.varunchoraria.com/",
      "intro": "Essays on AI, GTM, management, careers, and building things.",
      "listing": true,
      "content": "Essays on AI, GTM, management, careers, and building things.\n- My agent found 9.09%. I found one session. (2026-08-03) https://www.varunchoraria.com/11-sessions-are-not-a-ux-audit/: My agent found a UX problem on my site. Dead clicks had affected 9.09% of sessions, above the 5% threshold. It filed a GitHub issue and told me what to inspect.\n- The models were fine. I was the bug. (2026-08-02) https://www.varunchoraria.com/the-models-were-fine-i-was-the-bug/: I typed a command and nothing happened. No output, no error, no prompt coming back, just a cursor sitting there while my laptop got warm.\n- Who do we sell to vs. Who SHOULD we sell to (and how). (2026-08-01) https://www.varunchoraria.com/who-do-we-sell-to-vs-who-should-we-sell-to-and-how/: I’ve been a part of both startups and large organizations, and one consistent trend I’ve unfortunately seen is that ICP is considered a one-off exercise versus the continuous development it should be.\n- The Claude bill said $1,492. I paid $20. Here's how. (2026-07-30) https://www.varunchoraria.com/how-to-get-1492-out-of-a-20-claude-subscription/: I pay $20 a month for Claude. Last week I went digging through my own logs to find out what that actually buys, because I wanted to write an honest post about it and I did not want to guess.\n- Hunting Season for the Rest of Us (2026-07-24) https://www.varunchoraria.com/hunting-season-for-the-rest-of-us/: About a week into prepping for one of these companies, I could drop the category’s favourite piece of jargon into a sentence without flinching. Felt good about it too.\n- Github-ing (2026-07-16) https://www.varunchoraria.com/github-ing/: The world is moving toward systems-first work, and I do not think most people understand what that means yet.\n- How to design without AI Slop (2026-07-08) https://www.varunchoraria.com/how-to-design-without-ai-slop/: You are what you admire\n- I now have my own Master Shifu (2026-07-02) https://www.varunchoraria.com/i-now-have-my-own-master-shifu/: I’ve been collecting consulting casebooks for about a decade. Kellogg, Wharton, Ross, Darden, IIM (A, B, C, I, L), MDI, Drucker, McKinsey, Victor Cheng, Crack the Case — 19 books in total, some of them 300+ pages.\n- Nobody uses my MCP server. Building it still worked. (2026-06-28) https://www.varunchoraria.com/nobody-uses-my-mcp-server-building-it-still-worked/: Nobody’s going to read this, and I don’t mean this post specifically. I mean the whole thing: the blog, the about page, the work history. A hiring manager has 400 applications open and 11 minutes. A reader lands on one essay from a link ...\n- Maybe going around CIRCLES is worth it (2026-06-26) https://www.varunchoraria.com/maybe-going-around-circles-is-worth-it/: Frameworks get a bad rep. But early in your problem-solving journey, they’re scaffolding — not a crutch. Over time, as you get acquainted with more real-life scenarios, your intuition gets powerful enough to spot-check and work bottoms-up.\n- Too much parenting, too little adulting (2026-06-20) https://www.varunchoraria.com/too-much-parenting-too-little-adulting/: In the age of agentic AI, something I’ve first-hand witnessed as a people manager is that the default mental models are unchanged.\n- Killed By Google (2026-06-16) https://www.varunchoraria.com/killed-by-google/: I first came across killed by google and was honestly quite blown. I suppose the startup myth is real: your instinct is right 95% of the time, but your ideas are wrong probably 80% of the time.\n- Who owns what? (2026-05-17) https://www.varunchoraria.com/who-owns-what/: Everyone these days is asking the same question in the age of agentic AI: what should an AI do vs. what should a human do?\n- Team values I value (2026-04-25) https://www.varunchoraria.com/team-values-i-value/: We work with a bunch of vendors as a company of 44 people. As a manager, I have 2 choices:\n- I've been trying to get lucky for a decade. Here's what I learnt. (2026-04-19) https://www.varunchoraria.com/ive-been-trying-to-get-lucky-for-a-decade/: I think one of the things I have a perpetual obsession with is career optimization. It feels a lot like poker. There is no magic playbook, or a clean meritocratic path to it, at least not anymore.\n- Imposter syndrome as a first-time manager (2026-04-12) https://www.varunchoraria.com/imposter-syndrome-as-a-first-time-manager/: It’s been about a quarter since I became a manager. I’m wondering what value I’m actually adding to the organization, especially in the agentic era where skills, information, and knowledge are more or less commoditized. Here are my candi...\n- So you want to help your team succeed (2026-02-07) https://www.varunchoraria.com/so-you-want-to-help-your-team-succeed/: It’s been about a month since I’ve been a first-time manager. Before I started, I asked my CEO what advice he had for me. Here’s what he said:\n- \"You're not strategic enough\", said Manager (2025-10-26) https://www.varunchoraria.com/youre-not-strategic-enough-said-manager/: Cracks\n- 6 pro tips on communication that's sure to get you promoted (2025-06-15) https://www.varunchoraria.com/6-pro-tips-on-communication-thats-sure-to-get-you-promoted/: I’ve been in the workforce for almost a decade now, and I’ve observed a rather persistent truth: being a great IC gets you noticed, but becoming a great communicator is what gets you promoted.\n- What your career path eventually comes down to (2025-03-08) https://www.varunchoraria.com/what-your-career-path-eventually-comes-down-to/: I’ve spent most of the last 6 years trying to find the answer. Along the way, I’ve seen a pretty clear difference between the low and the top performers.\n- Problems tell you what to fix. Pain points tell you why it matters. (2025-01-30) https://www.varunchoraria.com/problems-tell-you-what-to-fix-pain-points-tell-you-why-it-matters/: Every marketer, founder, or sales professional starts with the same question: “What’s the pain point of my prospect?” But too often, we conflate pain with problems, and that mistake costs us deals, loyalty, and growth."
    },
    {
      "type": "page",
      "title": "leadership",
      "slug": "tags/leadership",
      "url": "https://www.varunchoraria.com/tags/leadership/",
      "intro": "I write about servant leadership, accountability, trust, team culture, and helping people succeed.",
      "listing": true,
      "content": "I write about servant leadership, accountability, trust, team culture, and helping people succeed.\n- Too much parenting, too little adulting (2026-06-20) https://www.varunchoraria.com/too-much-parenting-too-little-adulting/: In the age of agentic AI, something I’ve first-hand witnessed as a people manager is that the default mental models are unchanged.\n- Who owns what? (2026-05-17) https://www.varunchoraria.com/who-owns-what/: Everyone these days is asking the same question in the age of agentic AI: what should an AI do vs. what should a human do?\n- Team values I value (2026-04-25) https://www.varunchoraria.com/team-values-i-value/: We work with a bunch of vendors as a company of 44 people. As a manager, I have 2 choices:\n- I've been trying to get lucky for a decade. Here's what I learnt. (2026-04-19) https://www.varunchoraria.com/ive-been-trying-to-get-lucky-for-a-decade/: I think one of the things I have a perpetual obsession with is career optimization. It feels a lot like poker. There is no magic playbook, or a clean meritocratic path to it, at least not anymore.\n- Imposter syndrome as a first-time manager (2026-04-12) https://www.varunchoraria.com/imposter-syndrome-as-a-first-time-manager/: It’s been about a quarter since I became a manager. I’m wondering what value I’m actually adding to the organization, especially in the agentic era where skills, information, and knowledge are more or less commoditized. Here are my candi...\n- So you want to help your team succeed (2026-02-07) https://www.varunchoraria.com/so-you-want-to-help-your-team-succeed/: It’s been about a month since I’ve been a first-time manager. Before I started, I asked my CEO what advice he had for me. Here’s what he said:\n- \"You're not strategic enough\", said Manager (2025-10-26) https://www.varunchoraria.com/youre-not-strategic-enough-said-manager/: Cracks\n- 6 pro tips on communication that's sure to get you promoted (2025-06-15) https://www.varunchoraria.com/6-pro-tips-on-communication-thats-sure-to-get-you-promoted/: I’ve been in the workforce for almost a decade now, and I’ve observed a rather persistent truth: being a great IC gets you noticed, but becoming a great communicator is what gets you promoted."
    },
    {
      "type": "page",
      "title": "management",
      "slug": "tags/management",
      "url": "https://www.varunchoraria.com/tags/management/",
      "intro": "I write from experience about managing teams, giving feedback, building systems, and working with AI agents.",
      "listing": true,
      "content": "I write from experience about managing teams, giving feedback, building systems, and working with AI agents.\n- The models were fine. I was the bug. (2026-08-02) https://www.varunchoraria.com/the-models-were-fine-i-was-the-bug/: I typed a command and nothing happened. No output, no error, no prompt coming back, just a cursor sitting there while my laptop got warm.\n- Too much parenting, too little adulting (2026-06-20) https://www.varunchoraria.com/too-much-parenting-too-little-adulting/: In the age of agentic AI, something I’ve first-hand witnessed as a people manager is that the default mental models are unchanged.\n- Who owns what? (2026-05-17) https://www.varunchoraria.com/who-owns-what/: Everyone these days is asking the same question in the age of agentic AI: what should an AI do vs. what should a human do?\n- Team values I value (2026-04-25) https://www.varunchoraria.com/team-values-i-value/: We work with a bunch of vendors as a company of 44 people. As a manager, I have 2 choices:\n- Imposter syndrome as a first-time manager (2026-04-12) https://www.varunchoraria.com/imposter-syndrome-as-a-first-time-manager/: It’s been about a quarter since I became a manager. I’m wondering what value I’m actually adding to the organization, especially in the agentic era where skills, information, and knowledge are more or less commoditized. Here are my candi...\n- So you want to help your team succeed (2026-02-07) https://www.varunchoraria.com/so-you-want-to-help-your-team-succeed/: It’s been about a month since I’ve been a first-time manager. Before I started, I asked my CEO what advice he had for me. Here’s what he said:\n- \"You're not strategic enough\", said Manager (2025-10-26) https://www.varunchoraria.com/youre-not-strategic-enough-said-manager/: Cracks"
    },
    {
      "type": "page",
      "title": "Privacy",
      "slug": "privacy",
      "url": "https://www.varunchoraria.com/privacy/",
      "intro": "What this site measures, which services touch the data, and what I do not collect.",
      "listing": false,
      "content": "Last updated: 27 July 2026\n\nThe short version\n\nThis is a public website, not a product account. There is no login, checkout, newsletter form, or database of visitor profiles. I use analytics to learn which pages help and where the site is confusing. I do not sell personal information.\n\nAnalytics\n\nThe site uses Google Analytics and Microsoft Clarity. They can receive information such as the page you visited, approximate location derived from your connection, device and browser details, referral source, clicks, scroll depth, and session interactions. These services may use cookies or similar browser storage.\n\nI use this information to understand aggregate traffic, find broken journeys, and improve the site. Microsoft Clarity can create heatmaps and session replays. Do not enter sensitive information anywhere on this public site.\n\nYou can block analytics with your browser, an extension, or a consent and privacy tool. The core content still works.\n\nService providers and subprocessors\n\nThese are the service providers and subprocessors that can handle site or visitor data:\n\n\n  GitHub Pages hosts and serves the site.\n  Google Analytics measures traffic and on-site events.\n  Microsoft Clarity provides behavioral analytics, heatmaps, and session replays.\n  Cal.com handles scheduling only after you choose an external booking link.\n  LinkedIn, GitHub, and other linked sites apply their own privacy terms after you leave this domain.\n\n\nThose providers process data under their own terms and may handle it in other countries. I do not control their retention periods or independent use. This site does not offer a business product, so I do not offer a data processing agreement.\n\nContact\n\nFor a privacy question or deletion request tied to information I control, use the contact page. I may update this page when the site or its services change."
    },
    {
      "type": "page",
      "title": "product-marketing",
      "slug": "tags/product-marketing",
      "url": "https://www.varunchoraria.com/tags/product-marketing/",
      "intro": "I write about positioning, messaging, customer pain, product launches, and go-to-market strategy.",
      "listing": true,
      "content": "I write about positioning, messaging, customer pain, product launches, and go-to-market strategy.\n- Who do we sell to vs. Who SHOULD we sell to (and how). (2026-08-01) https://www.varunchoraria.com/who-do-we-sell-to-vs-who-should-we-sell-to-and-how/: I’ve been a part of both startups and large organizations, and one consistent trend I’ve unfortunately seen is that ICP is considered a one-off exercise versus the continuous development it should be.\n- Hunting Season for the Rest of Us (2026-07-24) https://www.varunchoraria.com/hunting-season-for-the-rest-of-us/: About a week into prepping for one of these companies, I could drop the category’s favourite piece of jargon into a sentence without flinching. Felt good about it too.\n- Problems tell you what to fix. Pain points tell you why it matters. (2025-01-30) https://www.varunchoraria.com/problems-tell-you-what-to-fix-pain-points-tell-you-why-it-matters/: Every marketer, founder, or sales professional starts with the same question: “What’s the pain point of my prospect?” But too often, we conflate pain with problems, and that mistake costs us deals, loyalty, and growth."
    },
    {
      "type": "page",
      "title": "product",
      "slug": "tags/product",
      "url": "https://www.varunchoraria.com/tags/product/",
      "intro": "I write about customer problems, product judgment, prioritization, and building useful software.",
      "listing": true,
      "content": "I write about customer problems, product judgment, prioritization, and building useful software.\n- Maybe going around CIRCLES is worth it (2026-06-26) https://www.varunchoraria.com/maybe-going-around-circles-is-worth-it/: Frameworks get a bad rep. But early in your problem-solving journey, they’re scaffolding — not a crutch. Over time, as you get acquainted with more real-life scenarios, your intuition gets powerful enough to spot-check and work bottoms-up."
    },
    {
      "type": "page",
      "title": "sales",
      "slug": "tags/sales",
      "url": "https://www.varunchoraria.com/tags/sales/",
      "intro": "I write about B2B sales, revenue systems, customer conversations, enablement, and pipeline.",
      "listing": true,
      "content": "I write about B2B sales, revenue systems, customer conversations, enablement, and pipeline.\n- Who do we sell to vs. Who SHOULD we sell to (and how). (2026-08-01) https://www.varunchoraria.com/who-do-we-sell-to-vs-who-should-we-sell-to-and-how/: I’ve been a part of both startups and large organizations, and one consistent trend I’ve unfortunately seen is that ICP is considered a one-off exercise versus the continuous development it should be.\n- Problems tell you what to fix. Pain points tell you why it matters. (2025-01-30) https://www.varunchoraria.com/problems-tell-you-what-to-fix-pain-points-tell-you-why-it-matters/: Every marketer, founder, or sales professional starts with the same question: “What’s the pain point of my prospect?” But too often, we conflate pain with problems, and that mistake costs us deals, loyalty, and growth."
    },
    {
      "type": "page",
      "title": "strategy",
      "slug": "tags/strategy",
      "url": "https://www.varunchoraria.com/tags/strategy/",
      "intro": "I write about prioritization, tradeoffs, durable advantages, strategic thinking, and better decisions.",
      "listing": true,
      "content": "I write about prioritization, tradeoffs, durable advantages, strategic thinking, and better decisions.\n- Github-ing (2026-07-16) https://www.varunchoraria.com/github-ing/: The world is moving toward systems-first work, and I do not think most people understand what that means yet.\n- I now have my own Master Shifu (2026-07-02) https://www.varunchoraria.com/i-now-have-my-own-master-shifu/: I’ve been collecting consulting casebooks for about a decade. Kellogg, Wharton, Ross, Darden, IIM (A, B, C, I, L), MDI, Drucker, McKinsey, Victor Cheng, Crack the Case — 19 books in total, some of them 300+ pages.\n- Nobody uses my MCP server. Building it still worked. (2026-06-28) https://www.varunchoraria.com/nobody-uses-my-mcp-server-building-it-still-worked/: Nobody’s going to read this, and I don’t mean this post specifically. I mean the whole thing: the blog, the about page, the work history. A hiring manager has 400 applications open and 11 minutes. A reader lands on one essay from a link ...\n- Killed By Google (2026-06-16) https://www.varunchoraria.com/killed-by-google/: I first came across killed by google and was honestly quite blown. I suppose the startup myth is real: your instinct is right 95% of the time, but your ideas are wrong probably 80% of the time.\n- Who owns what? (2026-05-17) https://www.varunchoraria.com/who-owns-what/: Everyone these days is asking the same question in the age of agentic AI: what should an AI do vs. what should a human do?\n- Team values I value (2026-04-25) https://www.varunchoraria.com/team-values-i-value/: We work with a bunch of vendors as a company of 44 people. As a manager, I have 2 choices:\n- I've been trying to get lucky for a decade. Here's what I learnt. (2026-04-19) https://www.varunchoraria.com/ive-been-trying-to-get-lucky-for-a-decade/: I think one of the things I have a perpetual obsession with is career optimization. It feels a lot like poker. There is no magic playbook, or a clean meritocratic path to it, at least not anymore.\n- \"You're not strategic enough\", said Manager (2025-10-26) https://www.varunchoraria.com/youre-not-strategic-enough-said-manager/: Cracks\n- What your career path eventually comes down to (2025-03-08) https://www.varunchoraria.com/what-your-career-path-eventually-comes-down-to/: I’ve spent most of the last 6 years trying to find the answer. Along the way, I’ve seen a pretty clear difference between the low and the top performers."
    },
    {
      "type": "page",
      "title": "Uses This",
      "slug": "uses-this",
      "url": "https://www.varunchoraria.com/uses-this/",
      "intro": "A running list of the hardware, software, and everyday gear I rely on.",
      "listing": false,
      "content": "Inspired by [Uses This](https://usesthis.com/), this page documents the tools and software I use to get things done. Last reviewed {{ site.time | date: \"%B %-d, %Y\" }}.\n\nEverything I still pay for earns its place or gets cut. I run a one-person marketing stack on an 8 GB MacBook Air: terminal agents, Markdown files, and tools I can own or swap out in an afternoon. The GTM stack further down is a separate list, because knowing a tool and being subscribed to it this month are different claims.\n\n{% for section in site.data.uses %}\n{{ section.title }}\n{% if section.note %}\n{{ section.note }}\n{% endif %}\n\n\n{%- for item in section.items %}\n  \n    {% include logo.html logo=item.logo label=item.name %}\n    {% if item.name_md %}{{ item.name_md | markdownify | remove: '' | remove: '' | strip }}{% elsif item.url %}{{ item.name }}{% else %}{{ item.name }}{% endif %}: {{ item.note | markdownify | remove: '' | remove: '' | strip }}\n  \n{%- endfor %}\n\n{% unless forloop.last %}{% endunless %}\n{% endfor %}\n\n\nThis is also the stack I run client work on. If you want it pointed at your pipeline, [here is how I consult]({{ '/consulting/' | relative_url }}).\n\n## Questions people ask\n\n{% include project-faqs.html %}"
    }
    
  ]
}