Stripe’s AI Strategy: Build More, Not Less

23:53Beyond that, from a user's first standpoint, just there's so much more they want from us. We've heard from a lot of users that for Stripe Tax, they want global filing. and it took us a long time to get to, you know, U.S. filing on Stripe Tax, meaning that, you know, every state in the U.S. will just automatically file for you. It's hard, it's a hard problem because you are dealing with so many different jurisdictions that accept filing in different ways. And so it took us a long time to get there. Global filing was the next ask. You know, we just don't want to have to think about this in any market.

24:28And we built that in about a third of the time that it took us to get to U.S. filing. Wow, with much greater complexity. Exactly, much greater complexity. And so it's just like an infinite list of asks and opportunities like that. And so if you're just relentlessly focused on users, you're going to focus on that. We launched recently a product called Stripe Treasury. And it's an interesting moment in sort of fintech right now because I think there is sort of an unbundling of what banking means. And Stripe is not a bank, does not aspire to be a bank, but does aspire to provide your banking portal and to help you with global money movement.

25:09This is something that we've been asked for for a very long time. And so we built Stryte Treasury to allow our users to hold funds in dozens of currencies across many countries and just build what the banking portal should be able to do. And so we've focused on bringing this experience to users extremely quickly. And what we're finding in every case is that the timelines that we thought were the best doable two years ago are now being compressed again and again and again. And it's because the direction that we're finding from an agentic engineering standpoint is kind of akin to injection molding.

26:01You know, it's a bit reductionist. Yeah. But when you look at sort of the growth in consumer packaged goods and toys and things like that, a lot of it just exploded post-war. And it was because we got really good in the 40s at melting plastic. It's the injection molding screw that sort of melts plastic very evenly and sort of injects it into a metal casting. And that's sort of how we're doing code production at this point. We're creating the mold and the templates. You mentioned markdown files. Those are extremely important. You need your agents files in every repo. And as long as you create those patterns and you sort of understand the pattern of how you integrate with financial institutions, you can hand off so much of the work to agents and they just get it done.

26:46And all of the time you end up spending is on the back end at code review. But even then, agents are doing code review even better than humans did before. So just, again, compressing timelines, it just feels like a much bigger opportunity than optimizing cost structure.

27:01David George:Yeah. I think you may be the first person to compare modern AI-based software engineering to injection molding. So this is great. We're covering your ground. And I've actually been to an injection molding plant before. Yeah, yeah. That's pretty amazing. It's pretty amazing, especially when you think about the fact that it was probably 70 years ago that it was created. James Watson Hendry, I think was the guy. Yes, that's right. Yeah, yeah. I've been in Pennsylvania. Yeah, there you go. That's great. You know, as an investor in Stripe, a large investor in Stripe, another way that I think about what you just described is the opportunity to optimize cost is finite, right?

27:37David George:You said long versus short. Yeah, I really like. Yeah, yeah, yeah. Yeah, no, it's optimizing cost structure is going short, sort of your own future potential, and just building faster, building more is going long, you know, your future potential. It's like the capital allocation equivalent of like returning cash. as opposed to reinvesting cash. Yeah, yeah. I got it. Okay, I want to shift over to a new topic, which you guys are on the cutting edge of and which gets a lot of play, but almost in like a nebulous sense. So I'd like to go a little bit deeper on it with you, which is sort of agents engaging in commerce on the internet.

28:11David George:So what is the state of where we are today? What are the bottlenecks? And how do you sort of view a framework for how that could be done in the future? Yeah. Yeah. So we are, and we talked about the Opus 4.5 Cambrian explosion moment. We haven't had that in agentic commerce today. There isn't a just, there aren't a ton of canonical use cases that you just see repeated over and over and over again. And I think there's a few reasons for that. One is we're missing primitives. Right. Yeah. And just that's something we're very focused on. We created with Tempo the machine payments protocol so that services can indicate what needs to be paid and how you can pay for it.

28:57So if you're selling an image or a piece of content online, you can request it and get a 402 response back. And it just says, here's how you buy me. So that's a primitive. We think machines will want to buy from other machines. and there's a question of like, what should checkout look like for agents? It's still sort of an open question. Browser automation is getting better, so maybe agents should just be crawling through checkout forms and tell them out. I think there's still a lot of open questions. Yeah, that's the skeuomorphic version.

29:29David George:Exactly. There's going to be a native version, yeah. Yeah, I think that's right. I know that's right. So we're sort of at the missing primitives phase and we're sort of figuring that out. And then there's a sociological phase of just in what ways will agentic commerce be better than non-agentic commerce? And there's some very non-speculative ways in which it'll be better. Like just checkout pages shouldn't exist for humans, you know, with straight blink and, you know, shop or shop pay. And it's very easy to get through a checkout page these days, but should you even need to go to one or should you just be able to say, buy it on a product display page?

30:10I think so. I think just checkout pages will go away. That'll be a thing of the past. Yeah, I think they'll go away. And so that's like a sort of non-speculative, but maybe less ambitious form of agentic commerce. I actually think that's interesting

30:21David George:because they will go away for human users too. This is not just like only commerce. I think they'll just go away. Okay, so StripeLink, like you have like 400 million users on StripeLink and then ShopPay probably has some comparatively large number and then you can use those and just automate the whole process. Exactly. Yeah, we launched recently the Link Agent Wallet so that there's now just a Link CLI. So an agent can just sweep up link credentials and go use them with humans in the loop to say how they can use them. I think one of the places we are the most excited about agentic commerce actually is B2B.

30:53So we launched Stripe Projects. Stripe Projects is a way to scaffold apps. That's sort of the narrow definition of it. But the most exciting thing that Stripe Projects is, is a way to provision B2B services agentically. Yep. So you can go, an agent can go adopt Vercel for hosting and they can do it without you needing to go to Vercel.com and do anything.

31:18David George:This has actually informed our investment thesis in a few investments recently. Yes. They're developer tools. Yes. And we say, okay, assume that the agents are going to be the shoppers in the future. Exactly. Like, is this the one that the agents will want to pick? Yes. Exactly. Exactly. And that's a very good framework. You know, an agent can go adopt BrowserBase. Yeah. And we actually had a really great demo a while back at Stripe where we used BrowserBase live via an agent to fill out an NCAA bracket. That's cool. That's great. Which I thought was awesome because I have three brothers and, you know, we all love sports, but I'm always sort of too busy to engage in the family pools.

31:57And so it was so cool to just watch the agent, you know, pick up a BrowserBase session and, you know, fire up ESPN.com and do a bit of research and just like fill it out. How'd it do? I don't actually know. I don't think we ever sort of ran the evals on it. I suspect probably as good as humans. Probably better. Yeah, exactly. Probably better. And so I do think agents sort of adopting B2B or sort of B2C services that are more utilitarian will be very popular. And so today, again, as a primitive, we're just trying to make it easier for agents to adopt services. But I do think the sort of sociology will continue to evolve.

32:39You know, the long running tasks are not sort of, I don't know that we've sort of figured out exactly how to do them on the consumer side.

32:49David George:Yeah, it's not there yet. My, I was trying the other day, I was using a plug code to try to compose a song for my niece's birthday. Oh. Yeah, it was pretty cool. And, you know, in the past, without a sooner, I just like definitely wouldn't have done that. with Asuno, I might do that. You might do it with Higgs field as well. We were talking about them earlier. But I'm not sure that I want to go create a 1099 or 999 a month account with them. I just want the sort of micro consumption for that. And so, you know, along with the actual commerce primitives, I think there's these micro consumption APIs that need to start to exist.

33:29And browser rates we're talking about, they're sort of leaning into it. But I'm very bullish on, you know, all these services standing up you know, sort of, not necessarily anonymous, but just ephemeral or one-time consumption. And I think that will really unlock agentic commerce too because then you won't have to say, well, I need to pay these guys 10 bucks a month, these guys 10 bucks a month, you just want to say that.

33:49David George:You get accounts and all this stuff. Exactly. Just use the service and pay. Agent, do this, discover the services, your budget is$15. Go. My 10-year-old has been creating rap songs. So I'm very familiar with all of these services. Yes, yes. And by the way, you know, the quality is actually pretty good and he posts them on Spotify. But the execution maybe is not quite as good, but the lyrics are very good from AI and the music itself is pretty good. So yeah, I would welcome that. So the critical thing, I think, or one of the critical things that you just said is this concept of micropayments or microtransactions.

34:24David George:So this has been something that has been talked about probably since the advent of the internet as an opportunity. Like, why do you think now it could work? well agents really increase human agency i guess it almost sounds um almost sounds redundant but you know talking about these um these applications uh composing a song for your knees filling out your ncaa racket whatever else it's really hard to do without agents to take a lot of time you're creating all these different accounts and so i'm just very bullish on services leaning into saying you can use me ephemerally, you can use me in a very sort of lightweight way.

35:06And I think to make that work, you're just going to need to support microtransactions. I think the case against microtransactions has always been, well, you're trying to consume content and if you're trying to sell an article, you're always going to be squeezed between the subscriber business model and the free business model.

35:27David George:Yeah, like you want to capture the excess as part of the subscriber. Exactly. You capture the excess or you do it with ads. And just like one or the other is better than what you're doing. And I think that was probably true in the past. But as you give agents more complex tasks, you know, you want them to be these little sort of hummingbirds, like going around the internet, just slurping up a little data here, you know, pulling it over here, you know, some very transient ephemeral storage, doing a little compute over here and you don't want the individual human to have to think about what they're using.

36:01You don't want them to think about what they're using in that these services are secure, right? Right, they're bona fide but you don't want them to have to sort of create accounts everywhere. So I think microtransactions will just be necessary for that economy to exist, the sort of agentic economy. And then on the flip side, they're now eminently possible because of stablecoins. Yeah. and stable coins are today relatively unergonomic. So if you're a human, you know, you have to go jump through a bunch of hoops. But if you just give an agent a budget, they can easily, you know, take dollars, move into stables, or, you know, just use a shared, a stored balance and find a way to check out.

36:42And they don't mind the sort of back and forth. Yeah, of course.

36:45David George:Yeah, yeah, I totally agree with that. And look, we've seen that transition that's happened in a lot of software business models today, right? Like the predominant business model of selling subscriptions and seats. And, you know, we've now seen a shift to consumption. And I think, you know, there's probably analogies where, you know, the consumers are better off because they can get access to more stuff. And then the businesses can access more people, which they otherwise couldn't access subscription service. So my hope is that that, you know, that does come into play. So on the topic of stablecoins, so we've covered sort of how you guys build software, how you move so fast in an AI development world.

37:23David George:You know, we've talked about the idea of commerce, you know, agentic commerce. Talk about stablecoins. Like you said, there's a lot of friction in the process today, but just state of the stablecoin market and what you see the opportunity to be for the next call it five years. So I'm an infrastructure nerd, so I always will go there first. And stable coins are just a better platform for moving money than exists otherwise. I strongly agree. And better for a couple of reasons. There are certain countries that have rolled out really good payment systems. Most of them are nationalized. UPI in India.

38:01Actually, it's interesting when you look at these economies, the percentage of sub -$5 payments in India, I believe, on UPI is, I think, something like 86%, whereas in the U.S. cards, I think it's single digits percent. So, you know, there's some good, cheap, fast payment schemes that are nationalized. PIX, another one in Brazil, which is growing mediocrically. But you need a shelling point for the global economy, right? Right. Just what can we all agree on? And that's where crypto rails solve this political problem. Right. Just this platform works everywhere. And so if we just all used it, the world, the global financial system would work better, be faster, be cheaper, and so on.

38:46So if we all like went to sleep tonight and woke up tomorrow and we all held stables, it would just be a better global.

38:52David George:Less friction and less cost. Exactly. In the economy. Exactly. Now that might be one of those should work this way, may never, you know, work this way problems. And so we are sort of resolutely focused on trying to change that. We talked earlier about Stripe Treasury. And, you know, we decided to make stable coins native to Stripe Treasury. So you can just hold a balance in stables just like you hold a balance in USD or EUR, TPP or anything like that. So we believe that the opportunity for stables is just faster money movement, cheaper money movement, and more global money movement. Right now, you can be a Stripe user in, I can't remember what the exact number is, but around 60 countries in fiat.

39:40But you can be a Stripe user in stablecoins in, I think, about 150 countries. So just bringing more people into the online economy in a way that allows them to transact with the AI companies. and as the software economy gets more and more global, bringing more people into it is more and more valuable. You can build a powerful AI company with two engineers in Thailand just as well as you can do it in the US or as well as you can do it in Brazil. So for stables, for us, it is higher performance, more global.

40:16David George:Yeah, and I like the position that you guys are in as it relates to stablecoins just because, again, talk about meeting the market where it is. like people don't want to cut over full wholesale and just drop all their fiat and move over to stable coins. So I think the relationships that you have and the sort of comprehensive offering that you have allows for adoption to happen at the pace that the enterprises want to adopt at. Or startups. Yeah, yeah. And all of this works, you know, when you really lean into it, Felix Bogor, Felix, I think as they're called now, which started as a remittance company between the US and Mexico, and now has other corridors as well.

40:56They built it on stables. And a few years in, they're now between 5 % and 10 % of remittances along that corridor, which is the largest remittance corridor in the world. It's amazing. And so you think about how long it takes companies to move fiat money efficiently, how long it took initially for the wisest of the world to do this, let alone the Western unions before it. and, you know, these are great companies but to get to 5 % to 10 % in just a few years is amazing. That's remarkable.

41:28David George:Yeah. I'd love to have you talk a little bit about Tempo. Yeah. So, you know, obviously it's a big important project that you guys are in the center of. Talk about, you know, where that is and what the aspirations for it are. Yeah, I think this, I'll just play the infrastructure nerd card again for a second. There are great blockchains, you know, and it, I think there are, you know, there was no obvious need for another blockchain outside of payments. Right. And there's a whole lot of reasons why a payment-specific blockchain can work really well. You know, privacy, for one, because, you know, blockchains are generally public and you can sort of reverse engineer, you know, what's happening on blockchains.

42:13So I want to make sure that privacy is a first-class primitive, making sure that throughput is never sacrificed. blockchains are disproportionately used for trading and so when you see massive trading events happen you'll typically see you know performance degrade a lot transaction fees like finding a way to create a blockchain where the transaction fee is never going to spike because you know other blockchains have abilities important yeah exactly and they have you know floating sort of gas fees and so the the project there is just saying how do we move money you know as efficiently, consistently, and cheaply as possible.

42:52Still in the early stages of building out, getting a lot of great traction. You were working with companies like DoorDash, you know, making it sort of the default, but not only blockchain in Stripe and feeling very optimistic about it.

43:04David George:That's great. That's awesome. Last topic I would love to cover with you is just the sort of token economy. And so, you know, it sounds a little buzzwordy, but you guys are, you know, I think over 2 trillion of volume now. And I think, you know, you and I would probably share the belief that the token economy is going to be one of the biggest things we've ever seen. So, you know, what do you think is kind of the state of play for accessing tokens? You know, obviously there's a lot of work that gets done in first-party applications today. But, you know, what do you think is sort of the future state of how we all access models and tokens?

43:42David George:Yeah. Well, I love, I've always loved the word token. And the reason is that it just sounds like an approximation of money. Yeah. And increasingly, that's just what it is, right? We're seeing, you know, we talked earlier about, you know, free trial abuse, multi-accounting, and the attacks that we're seeing against users whose platforms are general-purpose token consumers. You know, you can do just about anything with tokens on Cursor, right? You do just about anything with tokens on Replay. So the attacks against these users are very sophisticated. They're very reminiscent of what we see in terms of people trying to steal money from Stripe users.

44:26And so there's this blurring that you see between tokens and dollars. And so for us, we think about wanting to help users move money, store money, send money safely, compliantly, and so on. and we now feel this mandate to do the same thing on tokens, right? Yeah, of course. To protect our users in the same way. And, you know, over time, I think this is only going to happen more, right? You know, we talk about some of these long-running agentic tasks and a lot of them will replace, you know, services. You'll just be able to say, you know, close my books and I used to pay, you know, a human in tokens for that.

45:07Now, or many humans, now I pay fewer humans or a human, or sorry, in dollars for that. Now I pay fewer humans or a human in tokens, right? Because they're augmented by agents. So we just want to make sure that moving between tokens and dollars is as seamless and safe as moving between dollars and euros. We're at the beginning of this journey, but we think it's going to be a big part of the future of Stripe.

45:33David George:Yeah, that's awesome. Yeah, it's super exciting. And look, personally speaking, I think you guys have a real right to play a role there because of the relationship that you have with so many companies since inception all the way through to enterprise, as you said. Yeah, and it's funny, we're always a little bit careful internally about talking about rights, and so we talk about mandates. That's good. Yeah, we feel like... Win it and then win it again. Exactly. We have a mandate to help users with this, and so that's why we're really leaning into it. We're also focusing even more on spend management for our customers.

46:07They're using Stripe Treasury, they're thinking a lot about how much they're spending and how to reduce their spending, how their spending ties to the revenue. And a lot of token usage is actually in product. So you see sort of two different types of token usage. One is to build things, and this is sort of the OPEX management side of token management. And then there's the product efficacy side of token management, which is how good is my token-oriented or sort of token-driven product? How should I shift between different models to make it the most effective? And so we're really thinking about how do we help users on both of these.

46:49David George:Yeah, for sure. I mean, certainly in the latter, this sort of, you know, position as an orchestration layer that helps you effectively get the most out of tokens and, you know, also maintain some form of control, I think has like never been more top of mind than it is right now. Yes, absolutely. Yeah, and it's interesting because there's a hypothesis that, you know, software will be severely commoditized. And it's plausible. You know, I think we're at the, let's say we're either in the singularity or creeping towards the singularity. And it's very hard to estimate what a future looks like where, you know, models are sort of recursively generating models and so on and so on.

47:32But we are seeing the exact opposite right now where, you know, software creation is exploding. Yes. Customers are monetizing faster than ever. And we talked about our 2026 cohort growing 50 % larger and growing faster than 25 cohort. That one being 70 % larger and growing faster than the 24 cohort. And so, you know, it's more software being created. It's being adopted faster than ever before. And then more Stripe products are being adopted than ever before. 11 Labs is using 14 Stripe products. I just sort of wish it were 11 in a way. But actually, I'm glad it's 14.

48:10David George:Just make it 22. Yeah, yeah, 22. I was close with Matty and he does everything in denominations of 11. So you've got to shoot for 22. Exactly. And, you know, there's a lot more to software than just building, right? There's the expertise about how abstractions should thread through your entire business and sort of keeping those up to date. And you have, you know, turnover inside of your company. And so you build a system internally and then it starts to sort of degrade and that person leaves the company and the expertise goes away. And so we think about just how can we help companies across their entire revenue stack, you know, manage cash, manage revenue, close the books faster, grow faster, globalize faster.

48:48And that opportunity, I think, is just larger than it was ever was before. Yeah, couldn't agree more.

48:55David George:So in closing, I wanted to just get your take. And this is, I don't want to get too deep into, you know, the role of humans versus AI. but one of the things that Stripe has always been universally recognized for and appreciated for I think is your taste right and it sounds funny because you're an infrastructure company but I think it's generally agreed upon um how do you continue to uphold that at such a large scale and with so much of the work that you're doing now being done by AI yeah it's interesting how much the word taste is being used right now. And it's sort of hard sometimes. The cynical view is that it's how we're all justifying our future value.

49:37Yeah, that no one actually has taste, and it's just we're all reproducing. Yeah, exactly.

49:41David George:We're reproducing machines of things that we had previously perceived. Yes, yes. And my fingers are no longer as useful, but this nebulous notion of taste is, and models will never have taste, and so on, which is probably just not true. Yeah, I agree with you on that. But to your, you know, more proximate to your question, you know, product quality is really, really core to our culture and to our identity. And it's for a few reasons. Like one is just, we want users to have amazing tools that are carefully crafted that, you know, can make their, make them go faster, you know, enjoy, you know, the, the journey of building a company that much more.

50:25It's not an easy journey, but if you have tools that are surprisingly great, it feels that much better. So it's a big user lens. Another dimension of it is it's just more fun. You look at what you created with your team and you're like, wow, that looks amazing versus you kind of janked this thing together and it barely works. And it just isn't as fun to go through the slog the long nights, the long weekends, and so on, to build it. And I think the single thing I would say for scaling tastes, actually two things. One is I think it has to be a top step. You just have to say it over and over and over again.

51:08It's like the company strategy of win the startups and win them again. It's like talk about quality and talk about it again. And then the other thing is just use the product. Yes. And make it easy to use the product. in aviation. Simulation is so important. You can't just say, I hope the jet works. You have to prove to yourself that it does before you ever fly it. Or in NVIDIA, they really hit escape velocity was when they went from needing to wait for the fab to give them the chips to try the chips to actually simulate chip performance. And so we actually invest a lot in simulating the usage of products.

51:45Yeah, wow. And so you can point to an account and say, you know give me something that looks like that yeah but make it you know completely pii you know free you know randomize the growth rates so i don't actually know how big they are but just you know the day-to-day problems that they face you know the ups and downs of their business the seasonality you know give me a sense of of that and then you know make it feel live you know make it so that i'm feel like i'm receiving disputes from customers and refunds are happening and of course, these aren't real disputes. They're not real refunds, but just give me the experience that user's having so I can really live in it and use that with my team.

52:26And so I think, you know, you have the mandate to use the product and we really ask our EMs to lead this because they're the ones who control the resources and they're the ones who can sort of say, wow, this does not feel good. It needs to get fixed. So we're going to, you know, our next sprint is going to be elevating quality. And the other side, it's how do you give people the tools to do it more easily? So scaling taste for us is, you know, culture and it is the, you know, shibboleths and then it is the daily rhythms of just using the products and stepping into these issues. That's awesome.

52:58David George:That's so great. Will, this is so fun to talk to you about all these topics, you know, obviously covering how you're building so much product inside Stripe, which is, which is super unique and right on the cutting edge, I think. And the future of how we engage in agentic commerce and stable coins. You know, one of the things that I think you guys have said is, how do we assure that we build the next Stripe inside of Stripe? And I love that in closing, and I really appreciate the time. Yeah. Thank you. Great to be here. Thanks for listening to this episode of the A16Z podcast. If you liked this episode, be sure to like, comment, subscribe, leave us a rating or review, and share it with your friends and family.

53:41David George:For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow us on X at A16Z and subscribe to our Substack at a16z.substack.com. Thanks again for listening, and I'll see you in the next episode. As a reminder, the content here is for informational purposes only. It should not be taken as legal business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com forward slash disclosures.