Patrick Collison on Starting Stripe and What AI Changes

Patrick Collison remembered exactly where he was standing. A street in San Francisco, walking back from sushi with his brother John after attending Y Combinator's Startup School in 2009. They'd been kicking around the idea of building something for online payments. And then they just decided.

"We might as well," Collison told a stadium full of students at Startup School 2026, "because it probably won't be that hard."

It took two years to launch. Seventeen years later, Stripe processes payments for a significant share of the internet economy, and Collison was back on the same stage, this time being interviewed by YC partner Harj Taggar — someone Collison described as a co-founder from a previous company, which made for a notably unguarded conversation.

I'll be honest: I cover Main Street, not Sand Hill Road. A lot of what gets said at Y Combinator events isn't written for people who are weighing whether to open a second location or wondering if the AI scheduling tool their nephew recommended is worth the monthly fee. But this particular conversation had more in it for ordinary business builders than the setting might suggest. A few things in particular are worth pulling out.

---

What "know it yourself" actually means when you run a business

Collison made a point about AI and expertise that sounds technical but isn't. He described the difference between knowing something in your own head versus looking it up — or, now, asking an AI to look it up for you. His version used computer science terminology, but the underlying observation is simple: the knowledge you carry yourself is always faster and more useful than the knowledge you have to go fetch.

Think about what that means if you're running a shop or a service business and you're wondering how much to hand off to AI tools. Should you let it draft your customer emails? Handle your appointment reminders? Write up your invoices?

Probably some of that is fine. But Collison's point — and it's a fair one — is that the business owner who actually understands their customers, who knows from memory what their regulars need and how they like to be talked to, will always run circles around the one who outsources that knowing to a tool. AI can write a competent email. It cannot know that your longtime client hates exclamation points and always wants the invoice broken out by project phase. You know that. That knowledge lives in your head, and it's worth protecting.

Collison said he still writes everything himself — every communication, every note — because he finds the models' writing unconvincing in some way he can't fully articulate. "I've yet to send zero of those," he said, referring to the AI-suggested replies that Gmail and other tools now offer. Whether that's the right call for everyone is debatable. But the instinct behind it — that writing is thinking, and outsourcing one outsources the other — is worth sitting with before you let a tool take over your voice.

---

The question nobody asks before they start

The section of this talk that landed hardest for me was the part Collison called "what if you succeed."

He framed it as advice for people about to take on investors and employees: before you worry about what happens if you fail, ask yourself whether you'll actually want to do this thing for ten or twenty years if it works. He pointed to Larry Ellison still running Oracle after decades as an extreme example of the logic.

I've covered enough small business openings — and closings — to know that most people skip this question entirely. They spend enormous energy on the practical concerns: the lease terms, the startup costs, the licensing. They build a business plan around the first year. Very few sit down and ask: if this works the way I hope, what does my life actually look like in year eight?

The ones who don't ask tend to find out the answer the hard way. The restaurant owner who succeeds wildly and then realizes she hates managing staff. The contractor who builds a crew of twelve and discovers he liked working alone. The boutique that grows into a small chain and watches the owner who loved curation spend all her time on logistics. These aren't failure stories. They're success stories with a problem buried inside them, one that a single honest question — asked early enough — might have surfaced.

Collison's version of this was more optimistic. He said he genuinely loves what Stripe turned into, because working with businesses turned out to be more interesting than he expected. "I've never met a Stripe customer and thought that's boring," he said. Every business, in his telling, is a theory about how some part of the world works. He gets to see thousands of those theories tested in real time.

That's a specific answer to a specific question he actually asked himself. Most people don't get that lucky by accident.

---

The launch-slow story that isn't really about launching slow

Collison talked about Stripe's famously long runway before its public launch — roughly two years from first code to public availability in September 2011. He's right that this would have been unusual advice from any startup accelerator, including YC.

But the actual lesson isn't "take your time." It's something more useful: they had a real paying customer within two months of starting, someone whose needs they were building against in real time. The first customer, Ross Buché at a company called 28 North, could only charge a card at first. Then he wanted to see his charges. Then he wanted to refund a payment. Then — reasonably — he wanted to get paid. Stripe built each piece as the need arrived.

That's not slow. That's a different kind of fast — the kind where you're always solving an actual problem rather than a hypothetical one. The public launch was late. The customer feedback loop was immediate.

For a small business owner thinking about launching a new service or a side offering: you probably don't need to wait until everything is perfect, but you also don't need to announce to the world before anyone has used the thing. Find one person who has the problem you're solving, let them use your half-built solution, and let what they need next tell you what to build. That rhythm — real problem, real person, real feedback — is the part worth copying.

---

What Stripe's data actually says about right now

Collison shared something concrete, and it's worth paying attention to because it's grounded in real transaction data rather than vibes. According to Stripe's own numbers, new businesses are being started at roughly twice the rate they were a year ago — the largest year-over-year jump Stripe has observed. And the median new business, he said, is doing better financially than the median new business was a year ago. Time to first revenue is down. The probability of reaching meaningful revenue thresholds — a million dollars, five million — is improving across the board.

He was careful about the caveat: the future is hard to predict, and trends change. But speaking as of late July 2026, the data, in his words, suggests "there's never been a better time."

The reason, as Collison and Taggar both described it, is that established businesses are now genuinely willing to buy from small and new operators in a way they weren't before. The old hesitation — you're unproven, you might not exist in two years — hasn't disappeared, but it's been offset by a new fear: that doing nothing, sticking with old tools and old vendors, is now its own form of risk. That shift in buyer psychology is real, and if you sell to other businesses, it's directly relevant to you. This dynamic is part of what's shaping the AI-native company questions YC is grappling with across the board.

---

What I'd tell a Main Street business owner who watched this talk

Collison spent most of his time talking to people who want to build the next Stripe. That's not most of my readers.

But here's what translates: the concrete customer problem thing is real and it works at any scale. Stripe succeeded not because payments were an obviously great market, but because Collison and his brother had personally experienced how broken the existing options were. They built for a specific irritation felt by real people. If you're thinking about starting or expanding something, that's the test — not "is this a good idea in general," but "does someone I can name have this problem badly enough to pay me to solve it?"

The "what if you succeed" question is also real and it works at any scale. Before you sign the lease or take the loan or hire the first employee, ask yourself what your life looks like if this works. Not just in year one. In year seven, when you can't easily leave.

And the AI question — whether to use it, how much to use it, what to let it touch — is going to be with us for a while. Collison's answer is that he still does the things that require real knowing himself, and lets tools handle the rest. That's probably the right frame, even if the specific line you draw will be different than his.

The sushi was in Potrero Hill. It took two years to launch. Most of the things that matter happen in the middle, between the idea and the outcome — which is, as it happens, exactly where you live.

---

Dorothy "Dot" Williams covers small business and Main Street economics for Buzzrag.