Stripe’s AI Strategy: Build More, Not Less

**Will Gaybrick** (0:00)
If you want to ship more and build faster, you have to create a founder-like agency inside your company. A single engineer can do what two teams of engineers could do two years ago. Suddenly, the market opportunity landscape is just much broader, and you can do more with software.

**David George** (0:16)
How do you guys ship so much product?

**Will Gaybrick** (0:18)
Companies have skewed in that direction, where they've seen this new agentic efficiency and power as a way to optimize cost structure. Our belief, I'm being a little cheeky, but build everything?

**David George** (0:34)
Yes.

**Will Gaybrick** (0:34)
The Opus 45 Cambrian explosion moment.
We haven't had that in agentic commerce today. There's a few reasons for that. One is we are merging so much more code than last year and is stressing every system. So we create something called Stripe minions. You're not going to iterate, not go into planning mode, you're just going to say, this is what I want, go do it. That's where the world is going.

**David George** (0:55)
You guys are I think over 2 trillion of volume now. What do you think is sort of the future state of how we all access models and tokens?

**Will Gaybrick** (1:03)
I'm very bullish on.

**SPEAKER_3** (1:05)
AI is making engineers dramatically more productive. Stripe's response isn't to build the same things with fewer people, it's to build more. In this episode, a16z general partner David George sits down with Stripe's Will Gaybrick to unpack how AI is changing the way one of the world's largest technology companies builds. They discuss Stripe's internal coding agents, which recently generated 7,000 poll requests in a single week, why teams are becoming smaller and flatterer, and how Stripe is trying to give engineers the kind of agency traditionally associated with founders.
Then they look at what all this new software means for commerce itself. From AI agents sparring software from other agents to micropayments, stablecoins, and why Will believes the checkout page will eventually disappear all together.

**David George** (1:55)
I want to start with just a state of Stripe. So what is Stripe today? And when we originally invested in payment processor, but now I think the average AI company uses 11 different Stripe products. And at Sessions, I believe you had 288 distinct product launches. So I want to talk about velocity also, just to start with just what is the state of Stripe today?

**Will Gaybrick** (2:19)
So internally, we think about Stripe as having inverted our value proposition from being a payments company with sort of add-ons, to now being this multi-product platform where everything sort of focuses on financial infrastructure, helping you grow by reducing the friction and increasing the agency, to be more agile with your business model, to operate in more countries, and just go faster when it comes to everything that touches revenue and cash.
So just practically speaking, went from payments to then billing, subscriptions and invoicing, connect if you're a platform or marketplace, radar for meeting fraud, radar now for doing many more things than that, tax, just I think we don't actually count, but somewhere in and around 25, 30 products that are sort of headlining branded products, and of course, hundreds and thousands of features below that. But again, the framework we think about is really reducing frictions and increasing agency. So a good example, last year, we saw a lot of users for the first time experiencing free trial abuse, and this wasn't really an issue pre-AI because most Stripe users are software companies.

**David George** (3:33)
Yeah, high cost margins.

**Will Gaybrick** (3:34)
Yeah, exactly.

**David George** (3:35)
Joking burden wasn't there.

**Will Gaybrick** (3:36)
Yeah, so they're wasting a little compute, but it's negligible, it's minimus.
But now, software has a cost structure. And so, actually, I think Cursor was the first user that we were- I was going to say, it was our portfolio.

**SPEAKER_3** (3:48)
Yes, exactly.

**David George** (3:49)
We first experienced it. And I guess Internet users can be crafty, but yes.

**Will Gaybrick** (3:52)
Yes, yes. And I think it was something like one in six users of free trials were abusive. And so, you're just throwing money at these users who are just signing up for another account, another account, another account, maybe even doing model distillation and things like that. And so, we sort of got in the bunker with them and just stood up in a weekend, a pipeline where we were able to use our foundation model, look across the entire Stripe network, use our embeddings, and then after that, put a reasoning layer on top of so you could sort of say, we think this is a free trial abuser because, and point to those signals.