A founder showed me a demo last week that would have taken a team a quarter to build two years ago. He’d done it over a weekend. Alone.
It was very impressive. And yet it provided zero informational value on the viability of the business he was proposing.
That’s the strange place we’re in. AI collapsed the cost of building to near zero. You can go from idea to working prototype in a weekend — and so can the next person, and the person after that. When everyone can build the thing, building the thing stops being the point. But building was never the point! There’s an old adage that first-time founders focus on product, and repeat founders focus on distribution.
Velocity isn’t about the rate of shipping, it’s about pace of learning
Everyone talks about speed like it’s about shipping. Build faster, launch faster, iterate faster. We even say hustle is great execution meeting high velocity — so you’d think I’m all in on speed.
But when I say velocity, I don’t mean the rate at which you build. Especially in the earliest stages, I’d actually submit the velocity that matters is the pace at which you learn.
And learning is everything the demo doesn’t touch — who the buyer actually is, what they’ll pay, which channel reaches them, who to hire, why people churn.
AI compressed the build loop to almost nothing. It did not compress the learning loop. Nobody has invented a way to figure out your buyer persona in a weekend.
So the founders who win aren’t the ones who use the freed-up time to build more. They’re the ones who use it to learn more, faster — more customer conversations, more pricing experiments, more shots at figuring out distribution. Same clock, more cycles of the thing that’s still hard.
When I meet a founder, that’s what I’m underwriting. The learning rate.
The changing economics of build vs buy
One dynamic that AI does impact is the buyer’s build vs buy calculations.
Everyone’s got engineers messing around with Claude and Codex. They spin up an internal prototype in an afternoon and think — why would we pay a vendor for this? We’ll just build it. They are also facing pressure from management and their board/investors to ensure they have an AI strategy and that it needs to start showing up in operating results at some point.
I get the instinct. Right now, in this window, the perceived cost of building in-house looks lower than the cost of buying.
But anyone who actually ships AI products knows the catch. The models change under you constantly. Evals are hard. The weekend prototype is not the expensive part. Keeping it working as the ground shifts, and maintaining it, and trusting it, that’s the expensive part. And most teams haven’t hit that wall yet.
If you’re a founder selling into enterprises, this is your competition now. The customer’s own “we’ll just build it ourselves” reflex. Another version of this is a later-stage, better-funded startup telling the enterprise that they will send engineers to work with their in-house teams to build specifically for them. Startups have to think about their value prop relative to in-house teams AND better-funded companies offering custom work.
So where does the value actually land?
If building is free and the model is a commodity you rent by the token, the interesting question isn’t “is it AI.” It’s who keeps the margin.
At the app layer, the money doesn’t accrue to whoever has the cleverest prompt. It accrues to whoever owns the last mile — the workflow, the proprietary data that compounds, the distribution, the switching costs.
Everyone’s deck says “proprietary data.” Almost nobody has a data loop — the thing that gets better the more customers use it. Because building a loop takes time, and you need to stay alive long enough to let the loop actually work!
I spent years at NerdWallet watching this play out before AI was even the question. Remember that adage about repeat founders focusing on distribution? This is what it means in practice. We didn’t win because we had the smartest product. We won on distribution — getting in front of customers cheaper than anyone else. The product mattered. But the product was never the moat.
The honest version
AI made it easy to start and just as hard to matter.
The bottleneck was never building the product. It was finding customers who care, keeping them, and holding onto the margin — and none of that got easier. If anything, the crowd of people who can now build the same thing made it harder.
So the question I’d ask any founder isn’t “can you build it.” Of course you can. Everyone can.
It’s: can you learn faster than all the other people who can also build it? What is the hard-won insight you have about your customers and their needs that you’re building your roadmap around? What is the specific part of your pitch that makes prospective buyers’ eyes light up?
In some sense, business is simple — can you find enough people to pay you for solving their problems while still turning a profit? And yet it’s as hard as it has always been to find a scaleable, defensible business, to recruit and retain smart and motivated people, and to manage your own psychology while you do it!