MAIA JOSEBACHVILI: Hello, everyone. Welcome. Happy Sessions. I’m Maia, chief revenue officer of AI at Stripe. Twenty years ago, I founded Urban Escapes, an ecommerce travel and experience company—kayaking, whitewater rafting, skydiving. That’s me in white, 2007. And man, things were different back then. Now, this is before WeWork. So we thought we were being really clever because we rented out a conference room in a Midtown corporate building to save on rent. Now, back then, going from idea to a website where customers could pay online took months. I think about that experience a lot because if I was starting Urban Escapes today, I would basically do everything differently. And I know that because of you all. At Stripe, we work with the fastest growing AI companies in the world, many of them in this room. And we have a front row seat to what’s working, what’s not, and how the best companies scale.

Today, I want to share some of those insights because the playbook for how to run a company is being rewritten faster than any of us can wrap our heads around. So what are the four patterns we’re seeing? One: the top AI companies build faster. I think we’re all feeling that. Two: they sell globally by default. Your day one market is now the whole world. And three: pricing is evolving faster than most of us realize. Subscriptions were the breakthrough model 10 years ago, but not anymore. And four: they adopt to new go-to-market motions and fast. Building enterprise sales used to take a decade. Now it happens in year one. We’ll dig into each of these, but first, the big picture. We saw the top AI companies grow by 120% in 2025. So far, in ’26, by 175%. That’s nearly tripling in one year. The growth is not slowing down.

It’s accelerating. And the top ones are even more mind-blowing. Lovable reported $100 million in revenue in eight months. And then eight months later, $400 million. Cursor announced they hit a $1 billion dollar run rate in under two years. Three months later, $2 billion. Anthropic shared going from $0 in January 23 to $1 billion just two years later. And now they’re at a $30 billion run rate. It is wild. And it’s not just B2B. Our Link data shows consumer adoption of AI has doubled from just under six million to over 14 million consumers in just one year. And they’re spending more. Top Link buyers now spend $371 a month on AI. That’s up from $140 just one year ago. To put that into perspective, that’s more than the average American spends on internet, streaming, and phone service combined. People aren’t treating AI like a streaming service they might cancel. They’re treating it like a critical utility they can’t live without.

All right, so we have 175% revenue growth, adoption doubling, billion dollar run rates in months. What’s behind all this? Let’s dig into the four patterns, starting with speed. When I first started Urban Escapes, getting to a prototype took months. And, okay, I’m not kidding here. Building a shopping cart was so complex that in the first iteration, when you hit “buy,” we would tell the customers to mail a check to my apartment. And they did—like a lot of them—which is really unsafe when you think back to it. No, just like you no longer need months of engineering to accept a payment. You don’t need months of engineering to ship software anymore. And this chart really shows when that clicked. So, as late as 2024, iOS app releases were declining. Then agentic coding tools hit, and app launches jumped 24% month-over-month.

And Delaware incorporations followed the same pattern. You can see the huge spike when these tools went mainstream. Now, everyone’s talking about nontechnical founders vibe coding. But here’s something that’s really surprising. As startup creation has grown, the share of technical founders has increased by seven points in just one year. That means that a technical founder with AI tools can now do in days what used to take a team months. These are not side projects. These are real businesses. In fact, since embedding payments into developer platforms, like Replit or Vercel, we’ve seen each monthly cohort—which you see in these lines—go from idea to first charge faster and faster. And you can see it now takes builders less than six weeks to get their first paying customer. And I can promise you, none of these founders are asking people to send checks to their apartment.

All right, so what does this mean for you? Now, many of you are already doing these things, but for those that aren’t, here’s the playbook that’s emerging. First: obsess over developer productivity. Collapsing build times is a competitive advantage. The companies moving fastest, treat it as a priority, not as an afterthought. Second: build, sell, and iterate at the same time. That linear playbook, where you could finish the product and then go find your customers—too slow. The best teams do all three in parallel. And then third: when you’re posed with the “build versus buy” question, the fastest moving teams, build and buy. So focus your resources on what makes you most differentiated, and buy the infrastructure and building blocks to enable that. Now, getting to market fast is one thing, but where you sell matters just as much. With Urban Escapes, expansion was city by city—New York first, then Philly, Boston, DC.

And SaaS had the same logic, right? You’d explore selling in other markets, but the focus was really on nailing your home market. And then, once you had the playbook figured out, you’d hire a GM in either London or Dublin, you’d open an office, and then congrats, you were international. That playbook just doesn’t work anymore, and the data confirms it. A few years ago, the fastest growing SaaS companies would reach 25 countries in Year 1 and 50 by Year 3. But AI companies, in 2025, we saw 42 countries in Year 1. And, get this, 120 countries by Year 3. That means Kazakhstan is now showing up on the list for many AI companies. That was not on my bingo card. And this isn’t about selling to one person in these countries. There’s real revenue coming from them. Gamma is a great example. It’s an AI-powered slide and website builder.

They reported $100 million in revenue in their first year. And although they’re based in SF, the majority of their revenue comes from outside the US. Now, that’s not the exception anymore. It’s actually the rule. Across top AI companies, 48% of revenue comes from outside their home market. That’s nearly half of every dollar. Three years ago, that number was only 33%. So, global revenue isn’t a bonus anymore. It’s just the baseline. Now, let’s take a look at which markets are spending the most on AI tools on Stripe. Not surprisingly, we’re seeing strong spend in markets with high GDP, like the US, Japan, Germany. And then, there are some emerging countries that are coming up as hotspots like South Korea, Brazil, and India. The demand is definitely there. So how do you actually capture it? By meeting your customers where they are, which means offering local currencies and local payment methods.

Okay, get this. Localized pricing drives 18% higher cross-border revenue. And when you add at least one local payment method, we see more than a 7% conversion uplift. With these stats, it’s kind of crazy not to be doing this. A good check: put yourself in the shoes of a customer in Brazil. Can they pay in reais with Pix? If your answer isn’t, “Of course,” you’re leaving money on the table. The playbook for this one: three ways to pressure test your global readiness. Number one: localize prices, and make sure you have local payment methods in key markets. Second: automate tax collection because none of us want to get in trouble. And third: track revenue and conversion by country, and obsess over optimizing it. All right. Onto pricing, one of my favorite topics. Because the old models just aren’t working, and AI has upended what we know. So Urban Escapes—pricing, way more straightforward.

Skydiving: $300. Rafting: $100. Same experience, same price—done. And the SaaS corollary was similar with seats—same price for everyone. But as we all know, AI pricing is way more complicated. Value is elastic, and so is cost. Now, everyone might use the same AI tool, but they’re having vastly different experiences. Let’s take two users: one is an engineer who sets up a bunch of agents before bed and wakes up with code ready for a review. And the other is my mom, who last week told me very proudly—and I’m very proud of her too—that she replaced Safari with ChatGPT on her phone. One is creating tremendous engineering output, and the other has a very upgraded search experience. The value and the cost are different, and so the pricing should be too. Now, this change in value isn’t new. Every major technology shift has repriced software.

For on-prem, you paid once for all the code written up until installation. So a one-time license made sense. Then, cloud changed everything. Software became continuous and autoupdating, so recurring subscriptions made sense. Okay, so now is when you’d expect me to tell you the exact right pricing model for AI. Good news, bad news. Bad news first. We’re still in the early innings, and the gold standard hasn’t been stamped yet. But the good news is a clear pattern is emerging. So we’ll take Replit, a developer tools company that’s been around for almost a decade. When agentic coding took off, they pivoted their product, rethought their pricing, and hockey-sticked their growth. They started with flat subscriptions and then layered in credits as AI became core to the product. Subscriptions for predictability and then usage for value capture. The result? They’re targeting a billion dollars in ARR in Year 1 this year, and they’re not alone.

Usage-based pricing is fundamental to how AI delivers value. Two in three companies on the Forbes AI 50 have some form of usage-based pricing, and a majority of those companies are running a hybrid model, a subscription that anchors the relationship, layered with credits that scale with value. Now, knowing you need to be usage-based is one thing. Getting it right is another. There’s three steps to add to our playbook here. First: price in your customer’s language. Developers think in tokens. Enterprises, historically, in seats, but they’re getting increasingly much more comfortable with consumption. Match your pricing to the way your customers experience value. Second: show customers what they’re consuming before they get the bill. Who has ever opened their electricity bill and said, “Well, that’s exactly what I was expecting”? It’s always a surprise, and that’s not a great experience. Real-time visibility isn’t just nice to have. It’s a critical part of your product experience, and it’s how you prevent churn.

So practically, this means finding ways to tell your customers how much they’re using while they’re using. And then third: sell credits, not cost. Have your customers exchange money for credits once so they’re thinking about value, not dollars and cents, every time they use your product. Okay, onto the last pattern. Companies are adapting their go-to-market motions much faster. At Urban Escapes, we started with consumers first, and then believe it or not, built out an enterprise business because the Googles and Facebooks of the world loved taking their clients on adventures. And that’s been the norm. You start product-led, startups find you organically, and then you layer in enterprise sales years later—once you prove a product market fit at scale. I mean, Stripe followed the same arc. I was our first head of enterprise product, and that was just three years ago. Now, every AI founder I talk to is hiring a CRO.

Building out an enterprise motion isn’t a “three years from now” problem. People are doing it now. Cursor, the AI code editor, launched in ’23 as self-serve for developers. Then, they layered in sales-led to land enterprise contracts. Just a couple years later, they now have an incredibly impressive enterprise business that would have taken other companies decades to build. But that’s not the only shift. No, that would be way too easy. Entirely new motions are emerging, too. For one, channel sales have become a huge part of the AI go-to-market motion. A lot of people buy OpenAI or Anthropic through their cloud provider today. And now the AI companies are also building their own marketplaces so others can do the same. Channel sales is getting to be a much bigger part of the equation. And the billion- or really trillion-dollar question on everyone’s mind: “How do we all sell to our newest buyer, the agent?”

We’ve spent decades designing pricing around human psychology, anchoring on “good,” “better,” “best,”—$9.99 instead of $10. Agents don’t care about any of that. This chart is one of the clearest signals. In 2025, agent traffic to Stripe Docs 10x’ed in a single year. The purple line is humans. The pink line is agents. And you can see they’re converging. By the end of this year, agents will read more Stripe Docs than humans will. Now, here’s where a lot of companies underestimate the work. Go-to-market is the tip of the iceberg. Everything underneath it has to move with it. Adding a new go-to-market motion touches everything. Your product looks different for each motion because the onboarding flow is different. Pricing will have different nuances. You have commits for enterprises, sticker pricing for product-led. And then your org has to be built for all this. We all know that an enterprise sale is very different than how you do customer support for the long tail.

It’s not just a go-to-market change. We’re really talking about your whole entire business model. Now, most companies build revenue infrastructure the same way: one piece at a time. A billing tool here, a tax vendor there, payments bolted on when you need to. And when you’re growing slowly, a Frankenstein stack is much more manageable. But when you’re compressing a decade of go-to-market evolution into months, every single seam becomes liability. Here’s how I think about it: your customers, they’re channel-agnostic. The same customer might find you self-serve, but then graduate to enterprise and then have an agent implement a new product. They move across motions without thinking about it, and they certainly don’t think of themselves as a channel. Now, that means your systems have to move with them. The playbook for this one? First, have a clear graduation process. Know when self-serve becomes enterprise and how pricing changes with it.

Second: build unified systems. That means one customer object, one product catalog, and one data model. And third: design for agent readiness, where an agent could discover, evaluate, and activate your product without a human-in-the-loop. All right. We’ve covered a lot. And one thing is for sure: the playbook is being rewritten, from how you build on price to where you sell and go to market. The question we all have to ask ourselves is: “Are you driving this change or are you reacting to it?” And now I’m really excited to introduce someone who is very much in the driver’s seat, both as a founder of scaling his own company and as a platform enabling thousands of builders to do the same. Let’s welcome on stage Guillermo, CEO and founder of Vercel.

Hello. Thanks for being here.

GUILLERMO RAUCH: Thanks for having me.

MAIA JOSEBACHVILI: Great to have you.

GUILLERMO RAUCH: Great presentation.

MAIA JOSEBACHVILI: Thank you. You inspired a lot of it.

GUILLERMO RAUCH: Thank you.

MAIA JOSEBACHVILI: Okay. So we talked about a lot of things, but one thing’s for sure: the speed and the new types of builders that are emerging because of AI is pretty wild. And I think you have just been at the forefront of that. So share a little bit about how you think about it and what you’re seeing.

GUILLERMO RAUCH: Well, you talked about your idea started out with you wanted to create a website and make money. And one of the obsessions when I started Vercel was how quickly can you go from idea to software that’s global, that’s deployed, that’s secure, that’s performing everywhere on the planet. And the mission has continued, right? It started out with the developer experience. In fact, one of the early metaphors that one of my first angel investors used for Vercel was it’s a Stripe for deployments. And the thing that’s changed really that you kind of spoke to is nowadays I find myself thinking more about the agent as my customer, and the agentic developer experience. So the mission of Vercel has not really changed. It’s evolved. And instead of building just developer infrastructure, we’re building agentic infrastructure. And I think of the agent as my customer. I’m sweating the details of the error messages for the agent as my customer.

I think about the agent getting stuck. You were talking about Stripe, seeing all of this traffic coming from agents. It’s our responsibility at Vercel, and as developers, to upgrade the web for agents, which is a really exciting, I think, once-in-a-lifetime opportunity. In some ways it’s an upgrade. It’s going to get better for agents. But even technically sometimes it can be a “downgrade.” It’s just less HTML, more markdown. And as you said, it’s the early innings. So, we are figuring it out as much as you all are and sharing it in the open, and learning as we go, as well.

MAIA JOSEBACHVILI: When did you make that shift that the agent was the customer?

GUILLERMO RAUCH: As a founder, I spend a lot of time talking to customers, as much as I can. Sometimes on X, DMs, all these things. I started noticing there’s the data, of course—nowadays, 70% of page views on Vercel’s developer systems and documentation are coming from agents, which is nuts, 70%. And 90% of that is served as markdown. So obviously the data is telling us all the time. At one point, I remember I tweeted/Xeeted that “10%”—this is a while ago—like, “Oh my God, I woke up one day, and 10% of my sign-ups were coming from ChatGPT.” This is before agents. And those sign-ups, the data told us were more intentful. They tended to convert better, meaning that when you’ve been cooking with an assistant, the assistant understands your goals better. And so the data was telling us that, but the anecdata was people were coming to Vercel that were not as professional with software engineering. They were not as experienced. And they would speak to me in terms of what the agent told them. So they were carrying the message of the agent. I would find myself in conversation saying, “Can you just share the transcript?” It’s almost like, “Can I speak with your agent?”

MAIA JOSEBACHVILI: Absolutely.

GUILLERMO RAUCH: It’s actually building the software. And so I started noticing that the persona was changing, evolving, and true to our mission, more people are being able to build. And that honestly got me really excited.

MAIA JOSEBACHVILI: Yeah, that makes so much sense. And I think so many people are experiencing something similar.

GUILLERMO RAUCH: Yeah.

MAIA JOSEBACHVILI: Okay. So we talked a little bit about pricing, and I know that’s top of mind for everybody. How do you think… I mean, you guys have been at the forefront of usage-based pricing—

GUILLERMO RAUCH: Yeah

MAIA JOSEBACHVILI: And with agents, you really have to think about pricing differently. So what can you share with the audience about that?

GUILLERMO RAUCH: Yeah. Pricing is the journey of every entrepreneur, I guess.

MAIA JOSEBACHVILI: Such a journey.

GUILLERMO RAUCH: One of the things that’s always been tough about pricing is that you get, with this platforms, you get so many different kinds of customers. We have the customers that come in, and they’re building their first prototype, or the inkling of their first idea. And then we have the largest enterprises in the world that have mastered and studied the blade of cloud consumption. So they’re used to buying stuff on the go. They’re used to forecasting their cloud spend. They’re used to buying through marketplaces of hyperscalers. And so we’re always trying to find, is there a one size fits all? And to your point, I don’t think there is. I think—

MAIA JOSEBACHVILI: I wish. That’d be so much easier.

GUILLERMO RAUCH: I really wish it was just like, “Hey, one plan, fixed costs. Everybody go nuts.” In reality, it’s really nuanced. And so what we’ve tried to do over the last few years is, can we make full consumption models more digestible, more understandable? I think there is a huge edge if you’re building a business to really embracing the versatility of consumption because I love the slide where you were showing there is the mythical 20-agent-in-parallel engineer.

MAIA JOSEBACHVILI: Is that not true? I feel like that’s what you do.

GUILLERMO RAUCH: I have a few, but—

MAIA JOSEBACHVILI: I feel like that’s what you do.

GUILLERMO RAUCH: It’s true that you have the power users, right? And then you have the P75 that is using AI more occasionally. And I think over indexing on one or the other is really a mistake. And so what we found is, for example, real-time usage visibility. This is before agents. We realized if we want to make consumption scale for small businesses, we can’t just yeet, “Hey, here’s a CSV of everything that’s happened on the Vercel platform for the last three months. You figure it out.” So much of the user experience became the experience of understanding your costs, monitoring them, soft caps and hard caps in real time, beautiful graphs, forecasts, all of these things. By the way, I’m glad Stripe exists because it’s so much work, and, ultimately, it’s undifferentiated work. I think all of us—I think I speak for most of us in this room—we want to build great products, we want to build great agents, we want to ship. But I’ll tell you that, all of that work, for us, it really paid off. We went from customers that would get frustrated, like: “Okay, how can I understand this consumption thing?” To it really becomes a power, and it enables basically unlimited growth for their business.

MAIA JOSEBACHVILI: That makes a lot of sense. And I think the real-time visibility is something we talk a lot about, is how do we create that for people because I think that makes a big difference.

GUILLERMO RAUCH: Yeah.

MAIA JOSEBACHVILI: Okay. Switching gears: global. You guys have such a global customer base. Tell us more. Was that intentional? Has that just happened?

GUILLERMO RAUCH: Yeah. The first tagline for the company was: “Real-time global deployments.” I ditched “real-time” because no one really understood what it means for our deployment to be “real-time,” but the idea was you’re thinking in real-time, the software is getting deployed in real-time. And global was very important to me—I’m from Argentina originally, and I would travel from the US back to Argentina, back and forth a lot. The experience of the internet is not even across the world. I always said, “If I were to live in one place, it would be US-East-1.” Because when you go to Virginia, you go to the East Coast, they have a better internet there, especially if you sweat the milliseconds. So much of the internet lives in one region of the world, and I wanted to decentralize that. So Vercel today operates in 20 regions. By the way, for the nerdier people in the room, it’s still hard to pull data gravity and compute gravity out of US-East-1, but I think we’ve made tremendous progress.

We’ve made tremendous progress with static, dynamic hybrid deployments where we make some of the pages be more automatically cached near the visitor. We made CDN a default, and now there’s a huge opportunity to do the same for tokens. So Vercel has this product called the “AI Gateway.” You can think of it as one API key, access to every model. And one of our aspirations there is to decentralize token access. We have huge customers in Brazil, huge customers in Europe, that are serving their customers by getting their tokens from Portland and Virginia, which is pretty nuts. And the reason for that is that, today, we’re like in the early days of the internet. A lot of the premium compute that’s running the best models on the planet, we’re basically out of capacity all the time. We’re dealing with quotas and getting more capacity from the AI labs. But I expect that to change over time. Much like compute today is all over the planet, one of our goals is to create this token delivery network and make token access as close as possible to the end visitor. But the point remains that by designing to be global-first, you basically create this incredible upside for your business. And I think using infrastructure like Vercel and Stripe makes that a default.

MAIA JOSEBACHVILI: Well, you guys are doing a great job with it. Okay. Let’s try to hit on all the trends we talked about. Let’s talk about all the go-to-market motions. So you guys started very product-led as well, but you now have a wildly impressive enterprise business too. What’s that journey been like?

GUILLERMO RAUCH: Well, as I mentioned, it involved changing everything about the company, like adapting our pricing, adapting our product, our marketing, investing a lot more in security. But one of the things that we’re talking about a lot these days is agent-led growth. It’s not fully PLG, it’s not SLG, it’s agent-led. So I have incredible stories now of enterprises where there is a move by the CTO to start and sort of rethink the company from its bones. They create new initiatives to rethink existing codebases with AI. You might have heard rumors of Meta saying, “Look, our existing codebase is great, but agents are not so apt at using our old styles of writing code.” So we’re seeing this motion within enterprises where there’s a reinvigoration and rebirth of a lot of infrastructure thanks to AI. A lot of SaaS companies are thinking, “Okay, what’s the future of our business? Is it just seat-based? Is it doubling down on our UIs?” You might have heard of Salesforce announcing—this is kind of an important moment, I think, in the history of our industry—Benioff announcing the shipping of a CLI. You didn’t have that in your bingo card for 2026.

MAIA JOSEBACHVILI: Definitely not.

GUILLERMO RAUCH: What we’re seeing is this enterprise reinvention more towards primitives and building blocks, where enterprises are exposing themselves through APIs, SDKs, CLIs, MCPs. And what we’re seeing is there is tremendous appetite for AI to help accelerate this. So products like v0, which you have a bunch of data, a lot of enterprise data, but now, instead of procuring new software, you can just generate it. The go-to-market motion is accelerating. I think a lot of the C-suites that are making these decisions for the future of companies with AI, they are using the AI products themselves. I have champions within enterprises that tell me, “Well, I was using Claude Code over the holiday break. I now want to bring the power of Claude plus Vercel to my enterprise.” And so I think this world is at continuing to converge so much so that I think if your idea was, “I’m going to have a super differentiated feature set only for enterprises,” and you have to contact sales to even just imagine what the product is going to be like, I think you’re going to have a hard time because that agent wants to try out that feature right away.

MAIA JOSEBACHVILI: Totally. Okay. Last rapid fire question. We’ve gone through a bunch of trends. We shared some checklists. What’s one thing on your checklist that you think about all the time?

GUILLERMO RAUCH: I’m continuing to sort of obsess about the experience of the agent. Just this morning, I saw my agent getting stuck on a problem, and I really think about the cultural change that it’s going to take. All of us in this room, we were born into an internet that didn’t have AI, didn’t have agents. And so I think a lot of what’s going to make us successful in the future is this reinvention of every part of the stack, reinvention of the internet, reinvention of the web. My biggest piece of advice that I give to people is that like: “The faster you can get rid of your preconceptions—in some ways, a little bit of the ego of the things that the tools that you knew or the things that you thought you were good at—I think the faster we can actualize ourselves, continuing to ship, encourage people to build in public. I think a lot of the great things that have made companies like Stripe successful come from that, betting on the builders and the indie makers, and building in public.” So yeah, it was great to be here, and thanks for this for hosting me.

MAIA JOSEBACHVILI: Yeah. Obsess over the agent experience. I think that is a very clear takeaway, and one for all of us.

Guillermo, thank you. That was awesome. Before we wrap, I want to leave you all with this: I remember sitting around a campfire and a friend telling me to pursue Urban Escapes and that it was okay to quit my job to do that. I did. And it took months to build a website, two and a half years to get to four cities, and a lot of sweat along the way. And if I were that same person today—same campfire, same idea—I could have had my product live using Stripe and Vercel by the time that fire burned out. That’s what’s changed. It’s not just the tools. The entire starting line has moved. You can build in hours. You can sell in 42 countries on day one, and you can go from PLG to enterprise to fully agent-led before your Series A. So the question isn’t whether the opportunity is real.

You’ve seen the data, you just heard from G. The question is whether you’re moving fast enough to capture it. This is the most fun time to start a company, whether you’re a startup or one of the fastest-growing AI companies like Vercel, or an enterprise trying to navigate all of it. And our job at Stripe is to be here with you every step of the way. So go build, and know that Stripe will be here to help you as you scale. Thank you.