Patrick Collison: “What If You Succeed?”

The problem

Founders rehearse failure but not success

Every founder in a stadium like YC's Startup School has spent time imagining what happens if the company dies: the awkward conversation with investors, the team dispersing, the post-mortem blog post. But Patrick Collison, who has run Stripe for seventeen years and counting, says founders rarely ask the harder question in the other direction. You worry about whether you can get the thing off the ground; you rarely stop to ask whether you'll want to still be doing it in a decade.

The framing emerged naturally when interviewer Harj Taggar pressed Collison on whether Stripe's payments work — dealing with banks in person, paperwork in Latin, all the schlep — ever felt intellectually deadening for someone who grew up obsessed with physics and wrote a Lisp dialect as a teenager. Collison's answer reframed the question entirely: the schlep is a surface feature; what matters is whether the problem underneath is rich enough to keep you engaged at the scale that success demands.

This is not a small concern. The title of the talk, “What If You Succeed?”, is the literal question Collison says he thinks founders should sit with before they raise a significant amount of money — because at that point, the machine has a momentum of its own.

How they solve it

Four things Collison actually believes about building a company in 2026

The first answer is counterintuitive: know more, not less, even in an AI era. Collison uses a metaphor he calls cognitive L1 cache — borrowed from Jeff Dean's famous “numbers every programmer should know”. Fetching a fact from a model is like going across the network; knowing it in your head is like hitting L1. You can run far more cognitive “round trips” per second inside your own mind than you can by muttering into a voice interface or typing into a chat box. That premium on in-skull knowledge, Collison says, still shows up clearly in what the most talent-hungry companies pay for.

The second thing: the conventional lean-startup playbook — find a tiny crevice, buy some Google ads, validate, expand — may be losing its edge. Not because it was wrong, but because the internet is much larger and more thoroughly tilled than it was when Eric Ries wrote the book. In a world where AI makes it cheap to spin up organizations with many simultaneous capabilities, Collison thinks the better move is to aggressively decorrelate from the crowd: occupy territory nobody else is trying to occupy, up front. Anthropic, Anduril, and many of the decade's breakout companies are the opposite of lean startups.

Third: the founding moment matters less than the first production user. Stripe's own origin has become something of a legend — two brothers walking back from sushi in Potrero Hill after attending Startup School in 2009, deciding to start the company because, in Collison's telling, “it probably won't be that hard.” That turned out to be extremely wrong, but the point he draws from it isn't the drama of the moment. It's that they had a live production customer, Ross Buché at 28North, within two months of first writing code. Everything else — the two-year private beta, the slow ramp to public launch — grew from actual user feedback, not hypotheses.

Fourth, and perhaps most surprising: Stripe's actual transaction data is bullish on the current moment in a way that cuts against both the doom narrative and the centralization narrative. New businesses starting on Stripe are up roughly 2x year-over-year right now — the largest relative jump Collison has seen in any single year, larger than the pandemic bump of 2020. And critically, those new businesses are not just cheap AI side projects: the median new business is doing better this year than last, and time-to-first-revenue is falling. Collison thinks the reason is that incumbents, terrified of being left behind, are now genuinely willing to buy from startups in ways they weren't three years ago. The risk of the status quo has risen; that is good for anyone selling something new.

Takeaway

The quick version

• Ask “what if you succeed?” before you raise. Raising money locks you into a trajectory. Check whether you'd still want to be running this company in ten or seventeen years — Collison says that question is underdiscussed.
• Cognitive L1 cache still beats the model. Knowing things in your head lets you run more reasoning rounds per second than any AI interface. Companies still pay a premium for this; don't outsource your thinking too early.
• Decorrelate aggressively. The lean-startup niche-then-expand playbook is harder in a more crowded internet. Bigger, more contrarian bets may now have better expected value than iterating toward the obvious.
• The Stripe data says: now is the best time ever. 2x YoY new businesses, median outcomes improving, time to revenue falling. Fear of centralization is understandable but not yet what the numbers show.

“What if you succeed? You raise money, you have customers, you have employees, a whole thing — are you going to enjoy that? Are you going to want to work on that for 10 years, for 17 years?”— Patrick Collison, Stripe

Stripe was built on a sushi walk and a shrug. Seventeen years in, Collison's argument is that the shrug was the right posture — not because startups are easy, but because founders who ask the wrong questions in advance usually end up somewhere they didn't want to be.