In an era where AI agents are emerging as new participants in commerce, Stripe is redefining itself from a “payment company” to “the global financial infrastructure for agentic commerce,” countering the prevailing narrative that “AI = layoffs and cost-cutting” with “Build Everything.”

Article author, source: XSecurity

In August 2026, a16z general partner David George engaged in an in-depth conversation with Stripe’s President of Product and Business, Will Gaybrick. The title of this conversation was “Tokens Are the New Dollars.”

Stripe has evolved from a payment company into a multi-product financial infrastructure platform, while at the same time, a new generation of AI companies are growing and monetizing at an unprecedented pace. In this conversation, Gaybrick reveals a counterintuitive strategic choice: as AI dramatically increases engineer productivity, Stripe’s decision is not to do the same amount of work with fewer people, but to do more work with the same number of people.

What is Stripe today?

David George: I want to start from the basics. What is Stripe today? People still see it as a payment company. But you’ve clearly gone far beyond that. Maybe you could describe what Stripe looks like today from an outsider’s perspective.

Will Gaybrick: We began truly expanding beyond payments about five years ago. Today, we have a product suite called Stripe Revenue, which includes Billing, Tax, and Identity. These are all built at the edges of our core payment business. Customer feedback has been consistently clear: they want a simpler tech stack, fewer vendors, fewer integrations, and less to maintain. They start with us, then want us to solve their next problem, and the one after that. We do this because our customers asked us to.

David George: What in your product lineup are you most excited about? What are you building that hasn’t been released yet?

Will Gaybrick: These suites are still in relatively early stages. Tax is a great example—we currently handle sales tax calculations for half of the U.S. population. In the past, these were features that Stripe’s payment customers had to build themselves. So we feel very mature in payments, but in suites, we’ve just begun. As for what we’re building but haven’t yet released—I’m sorry, I can’t share too much. But I can say I’m very excited about the direction our product is heading over the next few years.

II. AI Productivity: A Growth Multiplier, Not a Cost-Cutting Tool

David George: I’d like to shift the topic to AI. How do you see AI changing Stripe’s products? How do you view AI’s impact on the way teams are organized? And what about AI’s influence on the future of internet commerce?

Will Gaybrick: First, look inward. We have a team called Stripe Agents, an applied research and engineering team that’s also productizing internal agent capabilities. They’re currently working on two main projects: one is Minions, a coding agent. It’s extremely powerful. Today, about30% of Stripe’s PRs come from Minions. I posted a tweet in January saying there were roughly 1,200 PRs per week at that time. Now it’s7,000 PRs per week. And these aren’t just low-quality code—they go through our full CI/CD and testing pipeline. Human engineers review all of these PRs.

David George: That’s incredible. 30% of the code came from the Agent. Do you think this percentage will continue to grow?

Will Gaybrick: I believe that, in the long term, the total number of software engineers will increase, not decrease. Because we are building more things, and overall, we need more engineers. Many people view AI efficiency as a way to optimize cost structures—you see layoffs, you see companies cutting operational expenses. But our belief is that this is an opportunity to build more. Optimizing cost structures is short-selling your own future potential, while building more faster is going long on your future potential. The best way to optimize cost structures is to grow faster.

AI is dramatically increasing engineers' productivity. Stripe's response isn't to do the same amount of work with fewer people—it's to do more.

—— Will Gaybrick, President of Product & Business at Stripe

David George: So you don’t believe the narrative that AI will reduce the number of engineers?

Will Gaybrick: I think there will be more software engineers. We need fewer engineers to do what we’re already doing, but because we’re building more, overall we need more engineers. You can look at the data from SaaS platforms. There’s a narrative that SaaS platforms are struggling. But our new SaaS platform pipeline grew by103% in 2026 compared to 2025. People are simply saying there’s so much room for SaaS platforms, and I’m building faster and faster. Moreover, the 2026 pipeline grew 50% faster than the 2025 pipeline, and the 2025 pipeline grew 70% faster than the 2024 pipeline. This is an exponential trend.

David George: This indeed refutes the claim that AI will kill software engineering. Have there been any changes in your internal team structure?

Will Gaybrick: We have an internal concept called "founder-like agency." If you want to ship more and build faster, you must create founder-level agency within your company. This means smaller, flatter teams with less managerial overhead. The new model isn’t about doing the same amount of work with fewer people—it’s about teams of the same size delivering three times the output. The product Stripe Projects received most of its PR from just one person—Scott Alexander. Now, with only one screen in front of him, he orchestrates 16 agents and moves at a much faster pace. This individual is essentially coordinating the parallel work of 16 people.

David George: So the engineer's role has shifted from writing code to orchestrating agents?

Will Gaybrick: This is more of an organizational-level change. On a broader scale, it’s not that these teams are making ICs more efficient; rather, these teams are now doing more. We have a culture around how to conduct product reviews and how to set quality standards—and we’re still doing all of that. The same people are now doing three times as much. The culture we’ve created, along with all the tools and norms we’ve built internally, are designed to help people build faster and better.

Three: Stripe Minions — from 1,200 to 7,000 PRs

David George: Can you tell me more about Minions? How does it work?

Will Gaybrick: Minions are one-shot. You give it a prompt, it builds, then automatically runs the full CI/CD and testing pipeline, and finally a human reviews it. No iteration, no need to enter planning mode. You simply say, “This is what I want—go do it,” and it generates a PR. We looked at January, which had about 1,200 PRs per week. Now that number is 7,000, and it’s growing every week.

David George: What do you think about the code quality produced by Minions?

Will Gaybrick: These code changes will go through the full testing process and be reviewed by humans. This isn’t like a toy project—it has become a core part of our engineering workflow. 30% of PRs come from the Agent, and the proportion is still rising. We’re also doing similar things elsewhere—using AI to help improve internal processes across Sales, Support, Risk, and nearly every department.

David George: Do you have a sales agent?

Will Gaybrick: We use it extensively internally. Different models, different tools, including models we’ve trained. However, I can’t disclose the scale of our GPU cluster. But I can say that we’re not just using AI—we’re also productizing AI capabilities. We have a product called Stripe Managed Payments for merchant-of-record services. If you’re a company like A-Riffs, you act as the merchant of record in your home market, but in long-tail markets—such as Kazakhstan and across Asia-Pacific—Stripe serves as the actual merchant of record. We need to do significant legal and compliance work for these products, and it must be done individually for each merchant and each jurisdiction. We use AI to accelerate these processes.

Four: From Startup to Enterprise Flywheel: Win All Startup Companies, Then Win Them Again

David George: I’d like to discuss how Stripe entered the market. How did you think about your market strategy?

Will Gaybrick: Someone summed it up very well. I think it was Colin at Clerk who said Stripe’s strategy is “win all the startups, then win them again.” That’s a great summary. We start with startups and stay with them as they grow. You see companies like DoorDash and Instacart—we partnered with them early on and grew alongside them. We also serve enterprise customers like Amazon and Microsoft. But startups make us better—they’re the most demanding customers, the fastest, the most demanding. They’ll tell you straight up: “Your reporting feature is terrible, fix it.” That kind of feedback is invaluable.

David George: Are you targeting from the enterprise side downward, or from the startup side upward?

Will Gaybrick: We start from the entrepreneurial side and work our way up. Startups are our starting point, and we grow alongside them. Today, our enterprise business is also substantial. But fundamentally, we are product-driven—we have a product-centric culture. We organize around product reviews, quality standards, and goal setting (similar to Google’s OKRs). We are not a sales-driven company, but we have also learned from Microsoft’s approach to organizing sales teams.

Five: Agentic Commerce: The Missing Infrastructure

David George: I’d like to discuss Agentic Commerce. How do you think payments and commerce will evolve in an agent-driven world?

Will Gaybrick: B2B is currently leading the way. Stripe Projects is a great example—it enables agents to autonomously set up B2B services. An agent can provision Vercel hosting for you without you needing to do anything on Vercel.com. That’s the most exciting part. But consumer-grade agentic commerce still lacks infrastructure. The biggest issue: agents don’t have wallets.

David George: The agent doesn't have a wallet. Can Stripe solve this issue?

Will Gaybrick: We have launched the Link Agent Wallet. There is now a Link CLI that allows agents to retrieve Link credentials and use them with human-in-the-loop oversight. We have also recently introduced the Link Agent Wallet, enabling agents to sweep up Link credentials and utilize them. Additionally, we have Project Network, which enables secure sharing of payment credentials and compliance qualifications across platforms.

David George: What about the checkout pages? Agents don’t need pages.

Will Gaybrick: I think the checkout page will disappear—not just for agents, but for human users too. With Stripe Link and Shop Pay, do human users even need a checkout page anymore? Or should you simply be able to click “Buy Now” right on the product page? I believe the checkout page will vanish. It will become obsolete. Link already has 400 million users. When payment credentials are securely stored and can be invoked with one click, the traditional multi-step checkout process becomes redundant.

One of the most exciting aspects of Stripe Projects is that it enables B2B services to be onboarded via an Agent. An Agent can be hosted on Vercel, and you don’t need to do anything on Vercel.com.

—— Will Gaybrick

David George: On a deeper level, how do agents trade with each other?

Will Gaybrick: We partnered with Tempo to create the Machine Payments Protocol, built on the HTTP 402 status code—a long-dormant status code originally reserved for future payment scenarios. When an Agent requests a paid service, the server responds with a 402 status code accompanied by pricing metadata. Any API, content service, or developer tool that implements the 402 response can be natively discovered and purchased by AI Agents, without any checkout process or human intervention. This means the infrastructure for Agent-to-Agent commerce is being directly built into web standards.

Six: Stablecoins: The Schelling Point of Global Financial Infrastructure

David George: I’d like to talk about stablecoins. What are your thoughts on this space?

Will Gaybrick: Stablecoins are simply a better platform for moving money than existing methods. If we all went to sleep tonight and woke up tomorrow with everyone holding stablecoins, the global financial system would be better—less friction, lower costs. Stripe has natively integrated stablecoins into Stripe Treasury. Today, as a Stripe user, you can operate with fiat currency in about 60 countries. But with stablecoins, you can operate in about 150 countries. That’s a massive difference.

David George: Why are Stablecoins better than traditional remittances?

Will Gaybrick: Think about how long it took companies to move fiat currency efficiently—how long it took Wise to get there, let alone Western Union before that. Native stablecoin companies achieve the same scale much faster. Stablecoins solve the Schelling point problem in global fund flows—in a world without stablecoins, cross-border payments require multiple intermediaries, currency conversions, and compliance checks, each layer adding friction and cost. Stablecoins provide a neutral, low-friction settlement layer.

David George: What about micropayments?

Will Gaybrick: I believe micropayments will finally become viable thanks to Stablecoins and Agents. It only works when you combine both—Agents capable of making micropayments, and Stablecoins making those micropayments economically feasible. You can imagine a future where Agents silently perform hundreds of micropayments on your behalf in the background, without you even needing to know. This is impossible in the fiat world due to high transaction fees, but it’s possible in the Stablecoin world.

If we all go to sleep tonight and wake up tomorrow with everyone holding stablecoins, the global financial system would be better—less friction, lower costs.

—— Will Gaybrick

Seven: Integration of Anti-Fraud and Token Economics

David George: How do you handle fraud prevention, especially new types of fraud brought about by AI?

Will Gaybrick: There’s this blurring between tokens and dollars. We have a product called Stripe Radar for fraud detection. The attacks facing AI platforms today—such as multi-account registration and free trial abuse—are becoming increasingly indistinguishable from attacks on financial platforms. These attacks are fundamentally economic in nature, just shifting the target from “stealing money” to “stealing compute.” We use foundational models, leveraging embedding vectors from Stripe’s entire network to identify bad actors, then overlay an inference layer to deliver interpretable judgments. After implementing this system, ElevenLabs blocks 2,000 free trial abusers per day.

David George: This sounds like more than just anti-fraud—it's a network effect product.

Will Gaybrick: Indeed. With each additional Stripe merchant, the value of this cross-network identity and behavior model increases. We now feel a mission to do the same with tokens—protecting our users in the same way. We want to ensure that moving between tokens and dollars is as seamless and secure as moving between dollars and euros. This is no longer just a payment issue; it’s about the flow of value.

David George: So is Stripe’s boundary expanding—from payments to all types of value flow?

Will Gaybrick: Our mission is to make internet commerce better for everyone. Whether that value is in dollars, euros, or tokens, we want these flows to be just as secure and seamless. It’s a larger market and a more complex challenge.

Eight: Scaling Taste: How Stripe Makes Product Decisions

David George: I’d like to discuss product philosophy. How does Stripe maintain product taste?

Will Gaybrick: We place great emphasis on product reviews. We have a set of quality standards similar to Apple’s approach, and we use an adapted version of Google’s OKR system. But I believe the most critical factor is maintaining small, autonomous teams. Founder-level autonomy means each team thinks like a founder—focusing on what users want and what delivers the best experience, rather than being constrained by layers of management approval.

David George: How to balance product consistency with team autonomy?

Will Gaybrick: Through culture, not processes. We share many values—such as an obsession with quality and a focus on user experience—but these aren’t enforced through rigid processes; they’re transmitted through hiring and daily product reviews. We’re very selective in hiring, ensuring everyone who joins shares the same standards. Then, during product reviews, we collectively examine designs and discuss details, allowing this shared taste to be passed down.

David George: The analogy with NVIDIA is interesting—they achieved escape velocity by simulating chip performance rather than waiting for fab feedback. Does Stripe have a similar approach?

Will Gaybrick: Yes, we also simulate the user product experience. We have an internal tool that simulates various user journeys to identify friction points. This is similar to NVIDIA’s simulation strategy—validating assumptions in a simulated environment before committing to actual development. This allows us to iterate faster, rather than waiting for user feedback to uncover issues.

Nine: Summary: Go Long on the Future

David George: One final question. If you had to summarize Stripe’s core belief about the AI era, what would it be?

Will Gaybrick: Build more, not less. When AI gives you ten times the productivity, you can use it to cut costs, lay off employees, and optimize operational expenses—that’s shorting your own future. Or, you can use it to build more, serve more customers, and enter more markets—that’s going long on your future. Stripe chose the latter. We’re not only using AI to make ourselves more efficient; we’re also building products that enable our customers to better leverage AI. From Minions to Link Agent Wallet, from Stablecoins to machine payment protocols, we’re building the infrastructure for a world where AI agents are deeply embedded in commerce.

David George: Very impressive. Thank you, Will.

Will Gaybrick: Thank you, David. That’s interesting. I agree that long-form conversations are the best format.