But what’s happening, and I can guarantee you this is happening in every room right now, is someone sits there and asks what’s truly unique about this, or what is this doing that other competitors are not doing. Because what’s happened with AI is there are so many competitors doing the exact same thing.

And, there’s the dreaded pause..

Michael Grinich said the exact same thing on X last week, where there will be a lot of niche companies who are gonna help with the AI transformation, and venture may not be the right funding model for them. This is one thing that internally we have given a name for, which is what I call the new valley: the five to twenty five million dollars in ARR. Or more commonly used, is this venture backable?

AI has fundamentally changed the game for software (and even some hardware) companies, where you can go from idea to revenue very quickly. That viral chart we kept seeing, everyone putting their logo on their own version of it. It just doesn’t tell us much anymore. And it’s getting crowded day by day. Every viral company that comes out has copycats in the next accelerator batch. So the stories all kinda sound the same. The why now is probably AI. It’s a large TAM. It’s a fragmented space. Lots of documents if you are in B2B.

But the more important question, which used to only happen in later stage rounds, is do we see a viable path to this company having multiple lines of revenue. Is the wedge they’ve gotten gonna lead them to adjacencies, each of which can be hundreds of millions of dollars? Or alternatively is their primary product large enough to make hundreds of millions of dollars (e.g. coding agents). Not every feature deserves to be its own company, and the ambition has to be worth the ten year journey. And, we have to decide on this with twelve (or sometimes six) months of data. That’s very, very hard to get right. But hey, that’s the job, as Dempsey would say.

Historically, there have been only two types of companies that have worked. One is the extremely fast, hype driven, blue ocean, massive global change like covid, remote work, or AI (no triple triple double double here). The other one is typically more measured, incremental even. The playbooks for how you run these companies are very different. The hype one is more land grab: underprice the competition, spend a lot, raise many rounds of funding. The incremental one is more figure out the details: customer loyalty, negative churn, one or two rounds, then profitable and fast growth.

Right now at the seed stage, every company is getting priced like it’s on the fast path. The fast story is the only one the market will pay for, and that’s the story founders want to tell. Collectively, we know that most of these companies are gonna be on the slow path, but nobody knows which one breaks out at the seed stage.

So that company we talked about at the start may not get funded. It may not die either. It just gets stuck, where what made you legible at the seed is now becoming the bar even at the later rounds. And, what’s painful especially from a founder’s perspective, is these companies ARE good. Customers are great, margins are good, founders are good. A lot of them are in unsexy markets where in the long term they might become amazing businesses. But the Series A squeeze IS happening and the middle is shrinking out. Carta found only 15.4% of Q1 2022 seed companies reached a Series A within two years. The 2018 cohort was 30.6%. This, in my view, will get worse every quarter.

Varunram nails how this discourse usually ends..

So what options do these companies have? There are three or four paths they can go down.

The first one is obviously a second act. Go back to showing a path towards how do you get to hundreds of millions of revenue from the first product. Some founders genuinely have it in them, and figuring out if you’re one of them is what the next two years of board meetings are about.

Second is M&A, which is happening a lot. Right now every incumbent needs an AI transformation. Salesforce, Workday, Meta are all acquiring companies. But they are primarily acquiring you for the talent, even though it’s an acquisition. It’s not the revenue. You spend four years building an $8M business and the acquirer values everything about it except the revenue.

Third is you stay. You kind of keep chugging along, build a good valuable company, and maybe potentially buy back the stock. Find alternative capital sources that allow you to enable the growth.

And the fourth, which is you take the bitter pill upfront. There’s a new generation of founders that are seeing this issue and opting out of the venture treadmill altogether, where essentially they seed-strap. Raise a little bit of money from friends and family, get profitable almost immediately, keep eighty+ percent of the company. And they are very open with investors that they’re never gonna raise again. They are the only honest people in this story. They still are able to find a way, even though for investors it may not be one of those truly venture backable companies.

And that’s the problem right now with the new valley. There’s a lot of new good companies, and these kind of good companies are everywhere now. But, venture capital is an outlier spotting business. Few companies return the entire fund and the other companies round to ~zero. And, founders know this math when they take our money. That was the deal. I don’t think there is a systemic fix for this from a capital perspective. Every year someone proposes something that’s more dividend like, where you pay the money back. Michael’s version was more like an SBIR for AI enablement. Indie.vc and Earnest Capital were the dividend like ones. Founders liked it, but LPs balked, because a capped dividend doesn’t fit any bucket they are currently set up to buy. The idea keeps dying on the capital side. Maybe this time could be different, but I don’t think so.

So going back to the investment committee meeting. When we see a unique insight, large ambition, a founder who’s found a path that’s unique, we lean in aggressively. Outliers never look obvious at first.

But there will be a lot of companies in the middle, good but not great, that we end up having to unfortunately decline. I can see a lot of other partnerships doing the same thing. And I don’t have a good answer for what happens to all of them. I’m not sure anyone does yet.

Thanks to Michael, Jake, Ethan, Joowon, Vishnu and others for reading drafts and providing edits!

He meant it as a joke. But if you’re raising this fall, that’s basically the game. In a consensus market most VCs aren’t doing independent work, they’re reading signals off you and comparing notes. The error of omission is just too high for funds to miss the hot deal. Whether your company is actually good comes later, sometimes much later. So with August fundraising season fully underway, here’s a brain dump of what gets read, and how.

Who you talk to at a firm really matters. Associates and principals are really great. I genuinely mean it. But look at what they’ve backed recently to understand their leverage and their standing in the firm. Newer, worked on a few great recent deals, understands your space? Talk to them, and get a warm intro to them too. If not, they might be on the way out. Figure out quickly whether the fund is actually interested or just giving you lip service. And know that once you’re assigned to someone in the CRM, you’re stuck with that person. Wars have been fought over deal attribution. You think I’m joking but no.

Eventually you have to get to the GPs, the capital P partners, the ones with actual authority and not just the title. You could cold email. But a warm intro changes everything. Angels are great for this, and a founder they’ve already backed is even better. They’ll run onto a plane to talk to you. Schedules vanish. You don’t get passed to an associate (usually).

Then the ask, and here’s where it gets tricky. Ask for too much and they look at you skeptically, unless you’re Elon. Ask for too little and you’re not ambitious enough. Too little dilution, it’s not worth it. Too much, it’s too good of a deal. There’s no science here, but come prepared with a resemblance of a plan for why you’re raising that much.

And the number you say out loud matters far more than you think, because you can’t take it back. You say you’re raising $30M. People say it’s too much. You have confidence you can do it. You end up not being able to. So you go back and say fine, we’ll take $20M. It doesn’t work that way. It seeds a doubt that you weren’t able to fundraise, which scares investors even more, and it shows poor judgement on why you needed that money in the first place. I always advise founders to aim a tad lower so you can actually raise the amount, ideally in a competitive scenario with multiple term sheets.

It’s still an art. A hard one.

On price, the number one thing founders don’t understand: VCs don’t want a “deal.” If the price is too good, it begets even more questions, like why is the price not higher. It’s weird but it’s true. Same with ownership. No established lead fund will do under 10% dilution, so don’t even ask. (Obviously there are exceptions, and if you have insane leverage, go for it.) The floor is set because someone inside that fund is showing their partners how this one returns the fund. And never compare on what your competitor got. It’s the single best way to tank the deal. You have no idea how that deal actually got done, though it’s good to have as a comp.

Your story gets read hardest of all, especially in this era where it’s easier to raise a seed and much harder to raise an A. Yoni Rechtman wrote this month that there’s no revenue number that gets you a Series A anymore. The bar is whether what you’re building is obvious, and “as soon as you’re justifying the size of your opportunity, you’ve probably already lost.” So share your unfair advantages in product, tech, or GTM (ideally all three). Ground them in your company, your people, what you’re observing. There’s probably a competitor out there already doing what you’re doing. Your unique insight is what gives a VC confidence on why you’ll win anyway.

People also severely discount how much hiring great folks gives you the Mandate of Heaven in a fundraise. You should hire great people for the company anyway, but if you have great new hires, showcase them in the pitch. I’m totally shocked how few companies do this. It helps VCs underwrite the downside, where worst case the company gets bought for the talent.

Small things get read too. Your traction slide should show the current month, even if you started fundraising last month. June numbers in August smell like stale goods. And when anyone asks about your timeline, the answer is two weeks. It’s always two weeks.

Remember you’re being read even when you’re not in the room. The valley, and New York especially, is insanely small. If you tell a fund you have a term sheet from someone, it’s literally a text away. VC is a co-opetition, everyone talks to everyone. It’s why all decks leak and news spreads like wildfire. Don’t lie, and don’t share anything you don’t want shared. Even if the upstanding VC keeps it close, another one won’t.

And vibes are really important, maybe more than any of this. Be you. Showcase what you’re excited by and highlight where you need help (we thrive on helping). You can play the silly game of negging and acting like someone you’re not, but trust me, it works far less than you think. Hype can buy you eyeballs and attention. It’s not the end all be all.

Finally, this market will read you top to bottom and still get it wrong. Even the best companies struggled to raise a round. Look at Anthropic. So if you’re not in a hot sector, strap in. Share what makes you a great contrarian bet, because everyone is trying to balance their portfolio and you never know where you’ll find a champion, even if you strike out with the top 100 firms. Figure out your default alive path. Valuation markers will come and go, and a brand and a big number are nice for the ego. What matters is fuel in the tank to keep your ambition high.

All it takes is one yes.

He said he’s going to think about it, same as he has for the last six months, which is why he hasn’t (smartly) quit his job.

I have this conversation all the time now. Everyone wants to start a company, same way everyone wanted to be a PM seven years ago. It looks shiny but it’s SO hard.

And you see a lot of news around how someone will build the one person billion dollar company, or how you never need venture capital and can bootstrap your way there. If that happens, great. I don’t buy it.

And sure sure, the commodity seller of money is pumping his bags. Guilty. But you have no idea how many conversations I have where I tell founders not to raise, especially when they don’t have a long term ambition or a 10 year mission or frankly any shot at a barbell outcome.

What bugs me about the one person company is the order. How many people you need used to be something you figured out later, after the mission and the roadmap. Now it’s the first thing people decide, and the ambition gets sized to fit.

And that’s the thing. Everything is downstream of your ambition. If your ambition is very large you’ll hire people. You’ll acquire companies. You’ll use capital as a weapon and not as a destiny. You’ll use it to accelerate your roadmap. You’ll push hard even with constraints. You won’t let capital be the limiter of your growth.

Even the bootstrapped companies everyone points to work this way. Turbopuffer hit $100M in revenue nineteen months in without raising a dollar of primary capital, the only money that changed hands was a small secondary so the team could take some liquidity off the table. Yet they’re a couple dozen people and hiring more. They didn’t decide to stay small, their ambition just outran their need for anyone else’s money. Meanwhile the labs building these agents are hiring thousands of people and raising tens of billions of dollars, which tells you something about whether the people closest to agents think hiring is over.

Also a word on keeping a high bar, because none of this means you lower it. Each person will matter even more when they’re managing a fleet of agents that can wreak havoc on your company. Hiring has never been more important. But hiring no one, I don’t buy it.

Might be famous last words. But, no founder with real ambition has ever asked me if the one person company is possible. They ask how to hire faster, or they want to buy the competitor that’s annoying them. Maybe the people asking about staying solo haven’t figured out what they want yet.

The engineer in New York doesn’t have the ambition yet, so he hasn’t quit. If he finds it, he won’t need six months to decide. He’ll be gone in a week and I hope I’ll be the first person he tells.

People letting life live them.

I kept thinking about people I know. Chained to high income tech jobs while secretly hating them. Just focusing on random intangible things instead of their true passion. Coasting.

Mode push comes from Formula 1. There are moments you push, moments you preserve your tires, moments you cruise so you don’t burn yourself out. Push is the gear the pit wall calls for over the radio. The driver holds it for a few laps and backs off.

It’s a gear most people never reach in their lives. They push when something forces them to, then settle back down. Inertia.

For a long time I thought the difference was discipline. Some people force themselves and some don’t. I don’t believe that anymore. The people in that gear aren’t forcing anything. They just can’t sit still. They’re constantly tinkering, constantly making something, and nobody is making them.

My industry thinks it knows why. VCs secretly want to back founders with a troubled childhood or a chip on their shoulder. The pain supposedly keeps the foot on the gas.

The first people I ever saw in that gear had none of that.

My mom would sit with us patiently, three boys, and teach us all the crafts she knew. She pushed us to learn music, prioritized time spent together as a family, and took us across India and the world young to show us different cultures. My grandfather left a comfortable life in Rajasthan to build his own business far away in Ahmedabad. My dad expanded that business tenfold and taught us the hard etiquette of doing things well and right. My grandmother was the steel, always making something: crochet, crafts, her work at the temple. My great grandmother lived to 95 and pushed to learn till the very end.

Both my brothers grew into builders themselves. To this day they’re my biggest supporters and the ones pushing me to a higher bar.

No one in that house had anything to prove. And yet nobody ever coasted.

Venture is a business of exceptions, and there are more ways to make an exceptional person than we admit.

I frankly still don’t know whether the gear comes from nature or nurture. I just know most people are waiting for a call to push.

I honestly didn’t have a good answer during the call but I have been thinking about it since then.

My honest reaction to launch videos is just.. nothing. Thumb scroll to the next tweet way before the three second attention grabbing preview. Watching these I guess has been part of a VCs job, but I have felt my own attention drop every time a new one hits the timeline. Same script: very dramatic, very expensive, doesn’t tell any resemblance of a story. No customer anywhere in it. Chest beating, portraying yourself as more than you are so people FOMO into the product, the round, who you are as a brand. Two minutes of film (or writing) whose real job is just boosting your valuation.

I have started calling all of this “sanitized slop”.

Luckily I think people are seeing through this. Two million views, eight comments. That ratio should embarrass someone. Paid partnerships on X doing the work attention used to do. VCs whose reach is bought, not earned, posting about authenticity. The human eye turns out to be very good at detecting slop. Nobody can articulate what’s off about any specific video, but the smell is there.

Some still escape the exponential, there’s always one great video a quarter. But polish stopped meaning what it used to. It used to signal money and taste. Now it mostly signals someone’s buying something.

What comes next? It goes retro. Back to what worked before any of this. Less aesthetic, more nature. Products out in the wild instead of a studio. Twitch style, in person. Ums and ahs. Typos. The customer in the picture instead of a founder explaining why they are the greatest thing since sliced bread. Show the outcome and disappear. The “human era”, basically. Authenticity wins because customer obsession is the one thing nobody’s figured out how to buy.

The box office is running the same experiment at a hundred times the budget. Obsession cost under a million dollars and outgrossed a new Star Wars movie this spring. Backrooms just became A24’s biggest film ever, and the director is a 20 year old who got his start posting horror videos from his bedroom.

So you still have to launch. Distribution matters, making noise to stand out matters, nobody’s exempt from telling the world they exist. But a launch is the last step, not the first. Find the customer who won’t shut up about you, point the camera at them and get out of frame. If you can’t find an obsessed customer yet, no refinements will help. Probably just go back to work.

The tldr from the one on Tuesday: “Hey man, I thought I was going to be the best thing since sliced bread, and then the market really humbled me. I have a term sheet with not the greatest terms, the fund is OK but it’s money in the bank. Should I bank this, or keep going?”

These are good companies. A few million in revenue. Manageable burn. But they just aren’t in one of the hot categories.

I have now gotten eight or nine of these calls in the last two months, all complaining about the exact same thing. VCs are mimetic. We follow in flocks into the hot categories. Right now that’s AI infra, defense, bio, or robotics. If you don’t have a little heat in one of those areas, we kind of ignore it. Founders think it’s a slight on their business. But it’s just not as interesting to us in the moment. (PS: each VC will claim to be different, including me who just led a Series A for a deeply unsexy market, but it’s still the truth)

I’m new-ish to venture but reading history none of this seems new. It’s been happening forever. Founders are just feeling it particularly right now.

So, the deal can still get done. It just won’t be one of the “hot” ones. Which means even at $3 million in revenue, your post is somewhere between $50 and $100 million. And if you can get a Series A term sheet at all, that itself is a huge win.

Numbers show it. Graduation rates from seed to Series A were once as high as 33% (in year 3), and for newer cohorts they’re trending under 20%.

But the founders are pissed. What the f*ck, man. I put in blood, sweat, and tears. I got my seed done easily. And now my Series A valuation isn’t even 2x or 2.5x what I raised at. Meanwhile some random AI company with $100K in revenue is raising at a $200M post. What is happening?

And the entire seed class is feeling it. This excellent essay from Nnamdi shows that the pool of active seed companies peaked in 2022 and has been shrinking ever since. Money is locked on AI, and perfectly good non-AI companies are quietly quitting.

That sucks. For you, the deal probably won’t happen the way you think it will. Or at least not at the valuation you think it will. This is THE bitter pill.

Taking it costs you. The SAFEs convert, the dilution piles up, and you own less than you think. A flat round feels like a loss, and you carry it into the next raise, where the first question might be why the valuation was so low. Any founder who says that doesn’t hurt is lying.

Here’s what I told him. Run a quick process with the established parties who are already interested. Then take it on clean terms IF it meaningfully accelerates your ambition. It’s money in the bank, and valuation is just a marker in time. The fuel to swing really hard, the ability to play offense instead of defense. That matters far, far more than whether you got the same terms as someone else.

I know what you are thinking. A VC telling a founder to take the money is as self serving as they get. Fine. But, I’m not pushing you into a round you don’t want. I’m describing the market as I see it on the ground, not the one that screams loudly on X.

And conviction matters most. A VC writing that check is putting money where their mouth is, not aping into the next $200M company in a hot category.

I tweeted last week that every venture-backed application company now has to be a data company or a fintech company, ideally both. This essay is the long version.

Let’s start with the shift in who is using the software. I wrote two years ago that per-seat pricing cracks once agents become the users, and we seem to have crossed the line: Cloudflare says agent traffic passed human traffic for the first time. This fact has been debated online, but the trend is clear that agents are going to be the primary customer for all software. Think about what this does to the business model. A thousand employees running a hundred thousand agents isn’t a hundred thousand seats. So, what can an application company charge for?

An agent leaves behind two things worth metering: the decisions it makes, and the money it moves. The decisions are data. The money is fintech. Those are the two companies you have to become.

Decisions

xAI has an option to buy Cursor for $60B, a company that’s now doing about $4B in annualized revenue. The software is NOT the main reason xAI had to pay up. Anthropic and OpenAI were already watching developers work in real time using Claude Code and Codex. Buying Cursor was the fastest way for xAI to get into the token flow. Musk (world’s first trillionaire mind you) said as much, that the record of how a million developers actually use models would go straight into Grok’s training, and the high price was the toll for skipping the years it would take to collect that data the slow way.

People rebuilt working Cursor clones within weeks when it first launched and none of them caught on, because Cursor won on taste. The thousand small calls about what to surface and when to disappear. A clone copies the interface but inherits none of it. It can never reproduce the years of those developers accepting and rejecting and rewriting what the model handed back. Cursor now trains its own models on those diffs. The product won on taste. The data, however, became its primary moat.

To see why those diffs are worth sixty billion, imagine replacing 90% of your employees with a team of geniuses who have no idea how your company operates. It’s just chaos. That is roughly what dropping a frontier model into your business feels like, and Fable 5 just made the problem more obvious. Because a model that solves 80% of real software tasks where last year’s best managed barely half is not the thing you’re short of. The geniuses are interchangeable, all brilliant, all hard to tell apart on any of these benchmarks.

They fail for one reason. None of them knows what the people you replaced knew.

The bandaid has been to pull that knowledge out of people’s heads and hand it to the model as context. But most of it was never in a structure you could empty out. It’s tacit, and it only ever surfaced in the choices people made. The deal they walked away from. The line of code reverted at 2am. That customer nobody chased, and nobody wrote down why. That’s the real stuff. You can’t write it down as workflows since a lot of it is judgment that is not being stored today.

To bridge this, we are now moving from a world of context → harness → judgment. Context was retrieval, the right pieces in front of the model. Harness was the scaffolding, the loop the model could run inside. Judgment is the last layer and the only one that compounds, everything left behind by every call and correction and reversal made on top of the data.

Every AI application pitch I see right now has the context slide as the moat. Context graphs, the why behind every decision, wire it all into the model. That part is table stakes now, because context is the one thing every competitor is assembling the same way.

The corrections are different. Think of them as a scorecard. Every time a user fixed what the model did, they recorded what right looks like in your business. That scorecard does two jobs nothing else can. It’s the training signal that tunes a rented model to your business. And it’s the test set, the only way to know whether your agent is actually getting better at the job, because no public benchmark measures your workflow. You don’t need to pretrain a model from scratch. Even Cursor didn’t. Its in-house models reportedly sit on top of an open-source base, with the diffs doing the differentiating. Fine-tuning and RL on top of frontier models got cheap enough that a Series B company can run this loop today. Two years ago you needed a lab.

Sarah Guo calls this territory the untrainable: work whose correctness can’t be scored from the outside. The corrections are how you come to own it.

The vertical AI leaders already run this play. Harvey is worth $11B and Legora past $5B, both selling into law, both racing past the standalone tool toward owning the entire matter, because the lawyer’s edits on a draft are the corrections nobody else gets to see. Rogo is doing the same inside finance, capturing how analysts actually build the model and revise the memo.

None of these companies trained a foundation model. They built the harness around a rented one and kept the judgment that runs through it. That’s the thing that compounds.

An incumbent like Figma owns more than SVGs. It has the history of how a design got from v1 to v47 and every version someone killed on the way, a graded record of design taste. Linear holds the argument under every closed ticket. Notion holds the shape a team’s thinking takes across a thousand edits. You can’t export any of this when a competitor tries to pull the customer off, and all of it is the answer a generic model doesn’t have.

Which is why the labs are buying judgment off the shelf. It started with human-labeled data: Mercor is worth $10B paying a network of experts $85 an hour. Meta paid $14B for Scale to own the pipeline. A startup in New York will now clean your apartment for free if you let it film the whole thing, because the robotics teams need to watch a human decide what to do next. And it’s led to many RL environment companies reaching hundreds of millions of dollars in annualized revenue selling this same judgment over long-horizon tasks.

The labs trained on the whole internet and ran out, so now they buy decisions directly.

Dollars

23andMe sat on DNA from fifteen million people, a dataset pharma would kill for in this day & age.. and still went bankrupt last year. If money doesn’t flow through your data, you are just funding a science project. Most founders are still sleeping on this half.

Toast figured this out years ago. A restaurant is basically a payment processor with a kitchen attached, and the payments make Toast far more money than the software running the floor. Ramp took it further. Free corporate card, no fees anywhere, a cent or two skimmed off every dollar of the hundred billion that runs across it. That’s a $32B company built on rounding errors. The free card was just the front door to the interchange, and the swipe fee holds because the network holds. Money even pays you while it sits, collecting float before it ever moves.

Not all money meters have a moat. One popular vibe-coding app reportedly makes about 50% margin on the credits it sells, most of its annualized revenue simply a markup on inference. But a token markup has no network behind it, and your own inference bill falls every quarter, so that margin melts as the models get cheaper. The durable fintech is the kind with lock-in underneath it: payments where the network holds, lending where the data underwrites a loan a bank can’t see.

Payment infrastructure for agents is now finally live. When an agent books the flight and orders the parts and pays the vendor, something has to authorize the charge, carry it, and take a cut. Stripe shipped a protocol for it, and Visa and Mastercard are racing to set the standard. OpenAI already skims a few points off whatever its agents buy. A trillion agents transacting is soon going to be the largest payments economy ever built.

Turn one into the other

The best application companies that will last stop treating the two halves as separate. Judgment is the record of decisions about the work. Fintech is the record of decisions about money. The strongest companies turn one into the other.

Shopify is the best example. It started as store software. Then it attached payments, then Shopify Capital, lending merchants money underwritten by the sales data already flowing through the store, loans a bank couldn’t underwrite on its own. The merchant grows, the sales grow, the data helps with the next loan. Roughly three quarters of Shopify’s revenue now comes from the money side, not the software subscription. You can take that data and offer products only you could offer. Stripe is doing the same with Radar, and Ramp runs the identical loop with spend data and the card.

Rippling is trying to do something similar. Its core object is the employee, so payroll and benefits and devices and the card all draw on one source of truth. It still hasn’t locked anyone out, and Gusto and Deel are growing right alongside it, but the company that owns the object compounds on it while everyone else still tries to assemble it by hand.

Guard the writes

But there’s the “headless” tension that no one has resolved. If all software will be used by agents, to stay useful you have to let the agents in, and to stay alive you can’t let them take everything. Every system of record is being asked to be open enough that an agent can plug in through whatever protocol is on offer, and closed enough that nobody migrates off once they have what they need. Salesforce cut Slack’s data off from Glean and the other outside agents this year. They’re just the first to do it out loud.

The way forward to build a durable application company will be a split. Let the agents read, since reads are cheap and important no matter what you do. Guard the writes.

The place where new judgment gets entered, where people and agents approve and correct and reverse each other. That is the part a rival can’t migrate, at least not easily. What they will scrape is yesterday’s state. The decision being made right now is the only thing that stays yours.

What if Anthropic builds this?

You can argue that the labs are already in the token flow, so where’s the moat? Claude Code sees every command a developer runs and every suggestion they wave off, and ChatGPT watches more decisions in a day than your product logs in a year. If the lab already sees everything, what do they need you for?

My answer: what their tool sees is mostly generic. The same coding and writing every model sees, the exact stuff the labs are racing each other to commoditize. The rare judgment lives deep inside one company: how your hospital reads scans, which deals your firm learned to walk away from. None of it ever touches a lab’s chat box.

And the labs spent years telling enterprises they don’t train on their data. They mean it.. at least I hope so. The tacit knowledge moving through the model inside your product stays yours by contract. They see the trace go by, and they agreed not to keep it. You’re the only one allowed to.

The fintech half is the one they don’t want anyway. A lab will happily take your data. It has no use for your loan book, your fraud losses, or your money-transmitter licenses in forty states. So for your buyers, the data makes you worth buying. The fintech makes you hard to dislodge.

Cursor built the best data engine in its category, and a lab paid sixty billion to own it. That’s either your dream or your warning, and you won’t know which until it’s too late to change course. So build the two things that survive. Accumulate the judgment. Sit in the path of the money. That’s how you outlast the “what if Anthropic builds this” question.

Thanks to Ethan Ding, Hrishi Olickel, Christina Li, Tina He & Vivek Trivedi for reading drafts of this essay!

And then one pitch changes all of your priors.

This founder spent the first forty minutes telling me his core product would be a commodity. He opened with the chart showing his EV trending to zero by month thirty-six, the part most pitches try to hide. Then he walked me through what the product actually was, which was a way to sit inside a workflow that generated data no one else could collect. He had already picked the two customers he wanted first because their data, together, would make everything else compound. Six years on the customer side of this market, drawing the path before he knew it was a map. That kind of insight often takes years to build.

The strongest pitches carry a unique insight in the tech, the market, or the GTM motion. Two of those is the bar today, three is a home run, and one just isn’t enough anymore. The bar has permanently moved. Early revenue won’t carry you through this. A $50K ARR chart in a crowded category gets pattern-matched as noise.

Secrets get earned the slow way. You go deep into the problem and keep asking why. You get to the root of how a customer actually works instead of putting a band-aid on their existing tools. And you have to LOVE the customer. If you’re selling sales software, you’d better enjoy spending time with sales people. Call it obsession. The founders who uncover these secrets have their customers on speed dial.

But you don’t have to come from the industry to find one. Some of the best founders I’ve backed are generalists who picked a problem and studied it harder than the people inside it. They show up with an analogy nobody else in the room had made: the way SaaS unbundled enterprise software in the 2010s, or how the cloud transition reshaped database vendors. Reading history was their secret. They earned it by going deep on one industry and recognizing it was about to replay in another.

In this era of sameness, narrative matters far more than it used to. You can hear when a founder has been deeper than they’re letting on. Faked secrets fall apart inside four follow-up questions. You can hear the script underneath. Your viral launch video won’t save you there. The real ones get weirder and more specific the deeper you push. The best ones paint trillion-dollar visions and walk you through the exact steps to get there. You really can’t shortcut your way to it.

The best pitches don’t end when the meeting does. They have you thinking about it at dinner with your family, still working out why his second customer mattered more than his first and how those two together create the network effect.

Before your next pitch, remember: stories create the frame. Logic fills in the colors.

PS: If the title seems familiar, it is. I wrote a worse version of this two years ago. This is advice I keep giving founders, so I figured it deserved a real update for where the market is now.

At his last job there was a team of twenty handling this stuff, and most of the growth came from pushing notifications to a billion users who already had the app installed. He’d spent nine years inside that machine. It hadn’t really occurred to him that distribution was now something he personally had to figure out. You could hear it land on the call.

I posted this on X a few days later:

Every big-company manager replied to tell me I knew nothing and that this is how it’s always been done. The startup managers replied the other way, saying yes, finally, somebody said it out loud.

I left LinkedIn ten years ago for basically the same reason, except I didn’t have any words for it back then. I just had this itch I couldn’t have explained to anyone if they’d asked. The vesting was good and the lunches were free and my manager was someone I genuinely liked and I never canceled a 1:1. Underneath all of that, there was this autopilot I’d somehow convinced myself was just how I happened to be wired. It took me a decade and a few failed startups and probably a hundred founder calls before I could land on what the word actually was.

Back to the 1:1s for a second. They actually started for a real reason, which is that Andy Grove was running Intel in the seventies and the chip market would kill you if you got a single quarter wrong. He needed his managers making sharp calls fast with information from the people closest to the actual work. The 1:1 was a leverage tool for the manager and a communication channel for the company. Forty years of broken telephone later, it’s mostly turned into a wellness check where the manager performs some amount of concern and the report shows up with a pre-written agenda from the night before, and both of them walk out feeling like they did something even though neither of them actually did anything that mattered.

Once you start seeing this pattern in the 1:1, you start seeing it everywhere else too. Google’s free food was a recruiting weapon back when the engineer war was getting nasty, and it was also designed to keep you in the office a few hours longer than you would have stayed otherwise. OKRs were how you focused a giant organization around a few real bets that the leadership team had picked. Stock comp was supposed to align you with outcomes the company hadn’t actually earned yet, on a four-year vesting cliff that conveniently kept you in your seat. Promo packets were supposed to keep managers from just promoting their friends and giving raises to whoever happened to be in the room. Every one of these solved something real at the time they were introduced. None of them were ever supposed to become a thing you were owed for showing up.

But that’s exactly what happened. An entire generation of tech workers came up believing the apparatus they walked into was the bare minimum any reasonable employer would provide, when it was really just a recruiting offer that nobody ever bothered to take back once they’d hired you. We built daycares for adults.

You don’t learn agency in a place that never asks you to have any. And ownership is the same kind of problem, in that you can’t really learn how to do it when there’s always somebody else around to do it for you. You learn it the first time you realize nobody is going to push a notification to a billion users to make your problem go away. You’re the only person in the building who can actually fix what’s broken.

You might be reading this and shrugging it off because it doesn’t seem to apply to you. You’re in a big company and you’re comfortable and the paycheck keeps showing up like clockwork. Maybe your manager really does care about you, and maybe the company really is organized in a way that lets the best ideas win. There’s only one way to actually find out for sure. Pick something a customer actually wants and chase it down without writing a doc or booking a kickoff or sticking it in your OKRs anywhere. Then watch carefully to see who in the room congratulates you for helping a customer and who reprimands you for skipping the process. That’s how you find out.

A designer’s job is to solve a problem for the customer. Design is the tool they use to communicate the answer. Same with others. We divided the work because each of us was generally excellent at one area. Then we mistook that “piece” for the whole job.

Solving problems is messy. The pain comes from the carrying. The travesty is sometimes being so close to the answer and then having to shelve it. You put it down half-formed and carry it through dinner, through bedtime, through tomorrow until something cracks.

That carrying is the work. The polished thing anyone else sees is what fell out at the end.

This is what the design-is-dead crowd keeps missing. Claude Design shipped this week. It does the output. It doesn’t sit with you in the brainstorm.

Gokul declared design the first AI casualty anyway: AI makes the output, so why have designers. It’s the cleanest version of a wrong idea I’ve read all year.

The designer isn’t going away. The designer is still the best person to solve the problem because they’re the one who’s been carrying it the longest. AI is good at giving you the output. The output is 0.1% of the work. The other 99.9% is the mess.

The thinking. The brainstorm. The wrong turns. The four versions you kill because they feel cheap. The morning you wake up with the answer because you’ve been sleeping on it for three nights.

AI gives you the output. It doesn’t help with the carrying.

So the death isn’t of design. It’s of design’s last mile. The slice that was just file delivery. Same for engineering, same for PM. If your whole job was making the artifact and handing it off, that’s the part AI eats away. That piece was never really the job. It was simply the part we paid people for because we had to.

If you solve the problem end to end, from the customer to what ships, you’ll be fine. More than fine. The walls between lanes already came down. Most people haven’t adapted.

People who only ever made the artifact have a real reason to worry. AI does that part cheap.

If you’re someone who carries the problem all the way through, the job has changed. The 99.9% nobody could see was always the thinking, the messiness, the pain. That just became the whole job.

Intention was always the problem. AI didn’t solve it. It made it the only one that matters.

Two attacks in 48 hours on the most visible person in AI. Someone who has undeniably changed the world.

I keep walking around San Francisco and the energy is different in just two months. People lower their voices at dinner when the topic turns to work. They hedge when they talk about their companies. Everyone has some private number in their head about what they are worth if everything changes in the next eighteen months.

A founder I angel invested in has not updated his board in a few months. His product works, revenue is growing, but a new Claude feature does 80% of what he spent two years building. He does not know what to tell them. So, he has resorted to saying nothing.

YC founders refresh Anthropic’s changelog the way traders watch earnings calls. Every release is a potential kill shot. Anthropic added $11B in annualized revenue in the last month and OpenAI just closed $122B at an $852B valuation. These companies ship features monthly that erase startups overnight.

School admissions decisions landed last week. Parents I know who got lucky are already past the celebration and onto the next panic: can we still afford to live here in three years? Anthropic, OpenAI, maybe others will IPO soon. That money floods a housing market that is already broken. You are either on the rocket or watching it from the sidewalk.

Layoffs arrive in quiet monthly waves now. No headlines, just Slack messages. “Restructuring.” Everyone knows what it means. The agents are now doing what teams of people used to.

A recent NBC poll found Americans view AI less favorably than ICE. The guy who threw that Molotov cocktail was 20 years old. He was not angry about economics. He was simply terrified of extinction.

And yet the laptops in the Mission are still open on any given Tuesday. Agents are running overnight but the people who launched them are already onto the next problem by morning. The founder with the upside down cap table called me yesterday. He wanted to pitch his next idea.

The fear is the same everywhere. The 20-year-old with the Molotov cocktail and the founder ripping apart his roadmap at midnight are living through the same moment. One of them reached for a bomb. The other is optimistically thinking about what he wants to build next.

Build, don’t burn. That is the only way through this.

Comparison is how ambitious people procrastinate. It feels productive because it involves the same subject matter, the same industry, the same people. You think you are studying the game. “Benchmarking”. Meanwhile your actual work stays untouched.

Tracking competitors used to make sense. Who shipped the feature first. Who priced lower. Who hired the sales rep you wanted. Execution was hard, the race was about execution, and watching was a legitimate part of the playbook.

AI killed that logic. Forty minutes vs. six weeks. Anyone can build the obvious thing now, the market map, the landing page, the MVP that does what the customer said they wanted. When execution is free, there’s nothing left worth copying. The surface is drowning in adequate, and the playbook everyone’s studying is already a commodity.

What wins in the near term are secrets. Things you can only learn by going somewhere nobody else thought to look, a distribution channel that only works because you understand a specific buyer’s workflow, a technical decision that seems wrong until you’ve spent months inside the problem.

I see this split in pitches every week. One founder has the competitor slide memorized. Knows the Gartner quadrant. Can recite every feature comparison. I ask why Customer #3 almost churned and they blink. No depth, just the standard “churn’s within benchmark ranges.” They know the map but they’ve never been to the territory.

Then there’s the founder who interrupts my question because she’s been thinking about it for 400 hours and already knows where I’m going. Has a take on some pricing quirk in construction bidding that sounds obscure, shouldn’t matter, and turns out to be the main insight. She found a crevice nobody else thought to enter and just kept going.

I ask about competitors and she stares like I asked about the weather. She’s somewhere they can’t see, and she got there by chasing questions everyone else dismissed. Don’t get me wrong. She’s competition-aware, but she doesn’t consider them in the same game.

The founders still tracking competitors are playing a finished game. The best companies are icebergs, and you can’t build one by watching what others ship.

Even investors should not be immune. I stopped reading the same ten newsletters everyone else reads. Started cloning repos instead, deconstructing and reconstructing frameworks I didn’t know because I wanted to understand how they actually worked before the consensus formed. Learned more about where AI tooling is heading in three hours - much more fun than hearing the same three tropes from peers.

There will always be a superficial leaderboard. Funding rounds, feature launches, press coverage. Don’t let it define you. Play your own game.

“You’ve already made your decision. You just want me to validate it, right?”

Silence. Then a laugh. He’d been working on this thing nights and weekends for over a year. He had a family, good job, kept telling himself it was a hobby. But he just couldn’t stop. Every morning he woke up thinking about it. Every meeting at work felt like time stolen from the real thing.

That’s the pull.

You can’t justify it on a spreadsheet. You can’t even explain it to yourself half the time. But it won’t go away. Especially when you try to make it.

Lenny Rachitsky has 1.2 million newsletter subscribers now. Before any of that, he was on a trip to Joshua Tree, on psychedelics, sitting on a rock for three hours watching some kind of Buddha visualization with one phrase stuck on repeat: I have wisdom to share. Three hours on one rock with one phrase.

Nine months into writing weekly, he told a friend he didn’t understand why he kept doing it. That he should be focusing on his startup. His friend asked him something he couldn’t answer: how often do you enjoy something and people actually value it? Maybe follow that thread.

He almost didn’t. Most of us almost never do.

Everyone’s felt some version of this. A problem that nags at you. A market that looks broken in a way only you seem to notice. A skill you keep building even though nobody asked. If you can’t articulate the thing, how can it be real?

So, you wait. The timing isn’t right. You need more experience. After the next promotion. After the next vest. After the kids are older. Your family, your friends, they reinforce it because they love you and stability feels like the loving answer.

But, that thing keeping you up at night? The problem you think is so glaring someone smarter must already be on it? There’s a very good chance nobody is. The gap between what you see and what everyone else sees is not a reason to wait. Act on it. Most of my greatest mistakes in my career have occurred because of this gap.

The cost of ignoring the pull is time. Scrolling past someone else’s launch in three years thinking, I saw that. I had that idea. Nobody is ever haunted by the things they tried.

The pull doesn’t show up often. FOMO shows up daily. Impulse shows up daily. The pull lingers for months, years sometimes, and you can’t argue your way out of it no matter how hard you try.

If you’re feeling it right now, you already know what it’s about. Share it openly, invest in it, build something based on it. Today.

In one, founders are swapping tools by the week. They max out token limits before lunch and feel anxious about it. A solo engineer is now shipping what took a team of five last year. People refactor workflows every quarter and still feel behind. Everyone around me has welcomed AI into every part of their workflow (thinking AND shipping).

In the other, I sit across from Fortune 500 execs who still think of LLMs as a toy. VPs of Engineering who haven’t opened Cursor. And, these aren’t just random people on the street. These are people with decades of experience who have refused to update their mental models. Their priors froze when LLMs couldn’t count the number of Rs in strawberry. The models have improved every week since. The harnesses around them have gotten dramatically better. They wouldn’t know because they refuse to get their hands dirty.

It’d be one thing if these realities existed in different parts of the world. But, I’m seeing them in one place. At the same dinner.

Block cut 4,000 people last month. 40% of the company. Jack was blunt: smaller teams with AI tools do more and do it better. He predicted most companies will follow within a year. The stock jumped 24%.

Working at startups, the “10x engineer” was always a mythical being. That one person on every team who shipped while everyone else was still debating the right abstraction layer. We told stories about them like Usain Bolt. Aspirational. One per generation.

I think the world has not priced in the fact that AI has raised the floor. A human paired with AI can already perform at the 10x level. The mythical 10x didn’t get better. The floor just rose to meet them.

The talent market used to follow a bell curve. Most people clustered around average. Companies built compensation and leveling around that distribution. Show up, do solid work, get promoted on schedule. That curve is splitting in half. Every week the tools get more capable, and more people will be expected to perform at the new level.

The people who have truly embraced AI have figured out their job is to be a conductor. Hold all the context, the feeling, the texture in their head while using a fleet of agents to handle the execution. It’s like playing chess while making sure a dozen trains don’t run into each other, even though each train has its own conductor. You just need to hear when something’s off.

The other side still meets their OKRs. I was at a dinner recently and sat next to a VP of Product. 16 years at big tech companies. Runs a tight ship. When I asked what he’d personally built with AI, he listed things his team had done. His hands hadn’t been dirty in years. Built a career on reliable competence. Shows up, hits deadlines, keeps the trains running.

That profile used to be the most valued in corporate America for half a century. $300K and a comfortable trajectory. The tools just caught up to it. Heck, let’s be honest, they are better than them.

AI should be renamed amplification intelligence. It simply showed who already had agency and curiosity, and amplified them. The variance used to always exist. Now, it’s clear as night and day.

This is coming for all of us faster than any of us like to admit. The two realities are colliding fast. And when they do, only one version of “good enough” survives.

Software companies borrowed the whole model. PM writes the spec, designer mocks it up, engineer builds it, QA tests it, marketing launches it. Stations on a line. The org chart grew around those skill gaps like a city grows around a river. Agile tried to fix it. We moved from waterfall to sprints but kept all the handoffs. Faster cycles, but the same stations.

AI collapsed those skill gaps in months. The org chart stayed exactly where it was.

I’ve been a generalist my entire career. Never picked a lane. At every company I just found the most important problem and went after it, regardless of whose job it was supposed to be. For most of my career that felt like swimming against the current.

Not anymore. No fast-growing startup I know hires just a PM or just a marketer anymore. They want tinkerers. That marketer? She’s in Claude figuring out what the last deploy changed, building the landing page herself, writing ad copy, and launching the campaign before she goes home. No tickets. No handoffs. The engineer ships the feature, creates the launch video, spins up agents to review the PR, and owns what happens after it goes live. Nobody told them to work this way. There was just a problem and they went after it.

Early startups always operated like this. No boundaries, just outcomes. The difference now is that AI lets you keep that scrappiness at 50, 100, 200 people. The gaps that forced specialization at scale are gone. These teams pay more in tokens than they’d spend on new hires.

I was at the office with a founder in our portfolio yesterday. He walked me through a feature. Customer research, design, code, launch. All him. I asked who else was involved and he kind of laughed. “Why would I hand it off? I had all the context.”

“Pick a lane” needs to be demolished. What to build is way more important than how to build now. Generalist vs. specialist was always the wrong debate. Ownership versus dependency. That’s the only question that matters.

Most people don’t cross though. I get it. Your title starts to feel like who you are. Touching someone else’s territory feels wrong. And yeah, the first few times you’ll be bad at it. But a rough version you ship with full context in your head beats whatever gets delivered six weeks later. Three handoffs strip out everything that matters. Nobody gives you permission to cross. Performance reviews actively punish it. We need to stop obeying a system that stopped making sense years ago.

Companies that don’t fix this will bleed their best people to ones that already have. Give them space, ambition, hard targets and watch what happens.

Ford’s assembly line was genius for 1913. The resistance between stations was real.

That resistance is gone. And every day you stay at yours, someone without your title, your permission, your experience is shipping the thing you’ve been waiting to hand off.

Economists call this revealed preferences. The gap between what you say and what you do. Every platform before TikTok asked users to build their own feed. Follow accounts, like posts, tell us what you like. TikTok skipped the asking. Dwell time over likes. Subtle, but important difference.

I’ve been thinking about how modern LLMs learn. There’s this technique called RLHF (reinforcement learning from human feedback). Human raters look at two outputs and pick which one’s better. Thumbs up, thumbs down, repeat a few million times. That’s how ChatGPT, Claude, Gemini all got less weird.

The problem is RLHF (typically) trains one model for everyone. Millions of preferences blended into one reward function. Your thumbs up and mine averaged together. The model converges toward what most people like most of the time.

What passes for personalization today doesn’t change this. Memory stores what you told the system. Your name, your job, your tone. Stated preference bolted onto the same model everyone else uses.

Tab, tab, tab. Every software tool throws off micro-decisions. Accept, reject, edit, regenerate, abandon. Code completions you take versus skip. Email drafts you send versus rewrite. Each one is a vote.

Unconscious, mostly. After a while you stop noticing. 2 AM, accepting code suggestions because you have token anxiety. Third rewrite on that email because your boss might read it wrong. Training data is the last thing on your mind.

Feeds wish they had data this clean. Scrolling is semi-conscious. You know you’re being fed content. But autocomplete? You don’t perform for autocomplete.

Right now, this data improves models for everyone. Cursor uses your accepts and rejects to make Cursor better for all users. Convergence.

It doesn’t have to work that way.

Imagine a version where the data makes your model diverge from mine. A model that becomes irreversibly yours. Your “AI” and my “AI” turn into different products over time because we used them differently, and the system noticed.

Last year I wrote about forking your tools. The power to consciously remix an interface. You decide to fork. Here the product forks itself. You didn’t configure anything. You worked, and your patterns became the product’s patterns.

The moat is the divergence, not the model.

Switching costs transform. Files travel fine between tools. Export, import, done. But the understanding doesn’t transfer. Thousands of micro-preferences the system learned that you never articulated. Switching means teaching a new tool from scratch.

I probably wouldn’t bother.

The infrastructure for per-user divergence isn’t here yet. But it’s coming. Capture the behavioral data now. Log the accepts, rejects, edits, regenerations per user. The companies sitting on that history will have a head start that takes years to replicate.

Users will feel it the moment they try to switch.

On the flip side, I talked to an exec who was proud of his AI rollout. Every team had copilots, dashboards tracked adoption, and he walked me through efficiency gains for twenty minutes. I asked how his hiring plan had changed. It hadn’t. He couldn’t name a single metric that moved because of AI, and headcount was still going up. He’d given every person in every role an AI assistant and left the structure exactly where it was.

Same dependencies between product and design. Same review cycles between design and eng. Same meetings with eight people so everyone “has context.” Product still says “prioritize.” Eng still says “capacity.” The board wants Q2 projections and something about “product velocity.” Everyone’s busy. They look around the room and it feels like progress.

These roles used to make sense. The PM had the idea, the engineer built it. That separation existed because you could not do the next step yourself. It was a real skill constraint, not some process thing. AI blew that up and nobody seems to have noticed, or at least nobody redrew the org chart.

At an event for non-technical folks using Claude Code, I watched an investment banker build a shopping app and push it live for customers that same afternoon. The distance between idea and done is basically zero now, for everyone, and most companies are still organized around a distance that no longer exists.

No fast-growing startup I know is hiring just a PM or just a marketer anymore. They want tinkerers. The person who talks to the customer in the morning, builds the fix by noon, and ships it before dinner. One brain holding the whole problem, start to finish. Not because they can’t afford to specialize. Because every time work changes hands, context leaks. And they learned to just not have that problem.

The best AI-native startups I see are spending thousands a month on token costs instead of adding headcount. Their biggest line item isn’t salaries. It’s compute. I think that says more about where companies are headed than any AI “strategy” deck.

Your company still has a role for every function and a gap between each one. Those gaps add up to weeks, and in those weeks your customers are already talking to someone who doesn’t work that way.

Every Fortune 500 is hiring a Chief AI Officer right now. Same instinct as “VP of Innovation” a decade ago. You created a role so the rest of the org could stay exactly as it is. Tobi Lutke doesn’t have a CAIO. He is the CAIO. His conviction comes from building, not briefings. Most CEOs would rather hire someone to have that conviction on their behalf, and it shows.

Startups have always been faster, that’s not news. What changed is that AI lets every person on a team take something from idea to shipped product without needing anyone else to touch it. Per-seat incumbents have the same tools on every laptop and a structure that forces the separation anyway.

The pot keeps boiling.

Nobody questions it anymore. Half the room is thinking the same thing. The other half are probably checking the progress of their agents. At a party.

All the parties are sober now. Young people don’t drink because they’re going back to work after. Not inspired by Bryan Johnson, although that’s probably a factor. The buzz they want now runs on tokens per day.

I keep noticing it on walks through the Mission. Laptops glowing everywhere. Cafes, sidewalks, heck even park benches. People walking with screens open like a flashlight guiding them somewhere. Less drunk laughter on the streets these days. More keystrokes.

Dinner conversations used to start with “what are you building?” That’s over. Now it’s “how many agents do you have running?” People drop the number the way they used to drop their follower count. Quietly competitive. The flex isn’t what you’ve accomplished anymore. It’s what’s working while you’re sitting here not working.

The vocabulary is what really gets me though. People describe models the way sommeliers describe wine. This one has better taste. That one hallucinates with more confidence. Opus is bold, Codex is smooth. They talk about harnesses and reins like they’re controlling horses. Invisible whips directing invisible labor. Someone at a dinner said they keep “Claude on a tight leash for code review but give it more slack for creative work.” We’ve started borrowing the language of how we treat animals for something none of us actually understand yet.

Waking up and checking what your agents produced overnight is the first thing now. Before coffee. Before texts. You open your laptop and grade homework you assigned in your sleep. Some of it is good. Most needs rework. But you start shipping a plan before you sleep just so you can wake up to more code written overnight. Saturdays became uninterrupted build windows. No meetings, no Slack, twelve hours of you and your agents. Sunday morning X is all terminal screenshots and shipping receipts. “What’d you ship this weekend?” replaced “what’d you do this weekend?”

The anxiety is rational, which is why it sticks. Every week some new benchmark drops that makes last month’s workflow feel prehistoric. Codex ships overnight processing. Opus gets faster. Context windows double. None of it reduces the pressure. It multiplies it. You can do more now. And someone already is. The window to be first at anything feels like it’s shrinking by the day. Literally, by the day.

I replaced Netflix with Claude Code. I lie in bed thinking about what I can spin up before I fall asleep, what can run while I’m unconscious. Reading a novel feels indulgent now. Watching a movie without a laptop open feels wasteful. This voice in my head that says “something could be running right now” just doesn’t shut off. I’m not even building a company. I’m just addicted to building my random ideas.

Everyone here knows they should step away more. That’s not the problem. The problem is what your brain does when you try. I still take aimless walks. The agents come with me now.

Last week, a Slack popped up. “Exciting news.” Your CEO is joining Microsoft. Most of the team is going with him. You are not on the list. You get paid, but the mission leaders have abandoned you.

In 2022, Adobe tried to buy Figma for $20 billion. Regulators decided to block it so Big Tech had to find a side door.

Microsoft paid Inflection $650 million in “licensing fees” and hired the CEO and most of the 70-person team. Amazon did the same with Adept, taking 80% of the technical staff. Google hired back the Character.AI founders for $2.7 billion, then repeated it with Windsurf and Hume. Same structure every time: license the technology, hire who you want, leave the rest.

Not an acquisition. No regulatory review. Everyone who built the thing doesn’t come along.

Now, founders are leaving their own companies. Last month, two co-founders of a $12 billion startup went back to the previous company. A year ago, a CEO left his $32 billion company for a competitor. When the people who started it don’t stay, why would anyone else?

In 2012, Instagram sold for $1 billion. All 13 employees joined Facebook. Founders stayed for six years. Everyone who took the risk shared the outcome.

There was an oath. Never written down. No founder signed it. No VC swore it in front of witnesses. But everyone understood: you do right by the people who bet on you. Employees who took pay cuts. Investors who wrote checks when nothing was proven. Partners who turned down safer options.

That oath is dead.

The week after your CEO announces, everything looks normal. Same standups. Same Slack channels. The people who got picked are negotiating their equity packages. Everyone else is quietly updating LinkedIn.

Why take the risk when the best case is watching your CEO leave without you?

VCs who funded your startup learned the same lesson. Five years ago, backing a competitor to a portfolio company was taboo. One fund forfeited a $21 million stake rather than create a conflict. Now mega funds back three, four, five companies in the same category. Fund sizes doubled. Companies stay private for 15 years. They can’t afford to miss a large category. So they bet on everyone and let the market decide. Watch them squeamishly talk about how “each” one is different.

I work at a boutique fund. We can’t hedge like that. We have to pick. I’m biased but that’s not the point.

The point is what we’re teaching. Every employee who got that Slack learns that “mission” is a recruiting pitch. Every founder who watches their VC fund a competitor learns they were a bet, not a partner.

I still back founders. I still believe there’s never been a better time to build. But I try to find people who still believe the oath matters. The ones who would rather go slower than abandon the people who believed in them. Even when faster often means richer.

So pick carefully. Ask your VC who else they have backed in your category. Ask what happened to employees at the founder’s last company. Understand what the numbers actually mean before you sign anything.

The oath is dead. But missionaries still exist. Finding them has never mattered more.

My timeline is calling it a disappointment. Ramp has been crowned the winner for a ~year now. $32 billion valuation, $1 billion “ARR”, all the buzz. Brex feels like a failure.

But Brex just put $5.15 billion in the bank. Real money. Everyone knows I’m a Ramp fan boy and Ramp’s $32 billion could absolutely be right. But we won’t really know until it clears.

Instacart raised at $39 billion. IPO’d at $10 billion. Klarna went from $45.6 billion to $6.7 billion in a single round. 85% haircut. The right corrections but TechCrunch called them failures.

Good reminder that the private numbers are not realized. We just like how it feels.

In this hype cycle, I am seeing founders pop champagne over term sheets that valued them at $2 billion on $5 million ARR. This high will last about a week. Then at the sign of the first downturn, the board will expect you to grow into it.

Meanwhile the stuff that actually matters - retention, margins, whether customers would notice if you disappeared - none of that makes the front page. Growth without retention is just a more expensive way to die.

Public markets don’t care about your Series D press release. They run the numbers. They look at churn. They discount the story and price the business. Sometimes brutally.

That leads to founders who are stuck. Raised at that $2 billion valuation. Can’t raise a down round because the optics are brutal. Can’t sell because the markup was fake. Can’t IPO because public investors would laugh. So they sit there, zombie companies with zombie cap tables, waiting for a market that’s never coming back. Or sell under the pref stack.

The VCs who marked those deals up have moved on to shinier objects. Still asking you to move fast and grow with abandon - while the founders are trapped.

Brex didn’t get trapped. Pedro and Henrique built something Capital One wanted to buy. They took the money. Their early investors made a fortune. Their employees got liquid.

So massive congrats to Brex on achieving a top 0.1% exit. If this is a disappointment, then I wish I get this kind of disappointment every day.