At the Stripe Conference, I saw the future of the AI economy.

Author: Gao Fei

Translation: Jiahui, ChainCatcher

In 1987, economist Robert Solow famously said: “You can see the footprints of the computer age everywhere, except in productivity statistics.”

This statement puzzled economists for nearly a decade. It wasn’t until the mid-1990s that the contribution of computers to productivity finally began to appear clearly in the data.

Today, in 2026, the same confusion is playing out around AI. Bubble talk rises and falls. Academic debates continue. Businesses hesitate. Macroeconomic signals remain blurry.

But in one place, AI’s impact on the economy is undisputed.

That place is Stripe.

In the past few days, I attended the Stripe Sessions conference in San Francisco. Stripe processes transactions amounting to nearly 2% of global GDP annually, with a total payment volume of $1.9 trillion, and over 5 million businesses on its platform.

In Forbes’ AI 50 list, 86% of the companies use Stripe. If the AI economy is a newborn, Stripe is the heart monitor in the delivery room. It records this baby’s heartbeat earlier and more accurately than almost anyone else.

A study released by the St. Louis Fed in early 2026 shows that AI-related investments have contributed nearly 40% to the US marginal GDP growth, surpassing the peak contribution of the tech sector during the internet bubble era. When these investments translate into revenue, a large portion of settlements happen on Stripe.

More importantly, Stripe is not just recording the heartbeat of the AI economy.

At this year’s conference, it announced a new economic paradigm: Agent Commerce, where agents become the primary actors in transactions.

In a group interview, Stripe co-founder and President John Collison said he expects that agents acting as buyers in commercial transactions will become mainstream within 12 to 18 months.

Two days. 288 product and feature launches. Over 10,000 attendees. A defining term: Agent Commerce. Here are my observations and reflections from the 2026 Stripe conference.

How fast is the AI economy growing?

Before discussing Agent Commerce, it’s necessary to look at the overall outline of the AI economy. Solow said in 1987 that computers’ footprints were missing from statistics. Nearly forty years later, AI is now clearly visible in Stripe’s data.

On the first morning of the conference, CEO Patrick Collison presented a set of data. Since the pandemic, the number of new businesses created on Stripe each month has remained high, but the growth curve was relatively flat. Starting from early 2026, this curve has shot almost vertically upward.

The most direct reason is that AI programming tools have lowered the barriers to starting a business. Many developers can now build billable products in days using vibe coding.

Patrick defined this as part of a larger concept: the entire economy is undergoing a platform reconstruction centered around AI.

Maia Josebachvili, Stripe’s Chief Revenue Officer for AI, added an external comparison: before 2024, the number of app releases in the iOS App Store was declining. But after AI programming tools emerged, app releases surged by 24% month-over-month.

This change is not only in quantity but also in quality. Stripe Atlas is one of the easiest ways for entrepreneurs to register companies in the US.

Last week, it celebrated its 100kth company. At the conference, I heard a staggering figure: companies registered via Atlas in 2025 at the same stage of their lifecycle generated twice the revenue of those in 2024. Meanwhile, the companies founded in 2026—just a few months ago—are already earning five times what the same period last year’s companies did.

In the AI economy report on the first afternoon, Maia listed several companies driving the rise of the AI economy.

Lovable achieved $100 million in revenue in eight months, reaching $400 million in the following eight months. Cursor hit an annualized revenue of $1 billion in less than two years, then doubled to $2 billion in three months.

Leading AI-native companies on Stripe grew 120% in 2025. By 2026, their growth rate has reached 575%.

Consumer-side growth is equally rapid. The highest-spending consumer group spends $371 per month on AI products—more than the combined monthly online, streaming, and mobile phone expenses of the average American. I roughly calculated my own monthly token expenses—they had long surpassed my mobile bill.

Patrick also made a comparison: the growth rate of companies on Stripe is 17 times faster than the global economy.

The next day, John Collison directly referenced Solow’s paradox and used a historical analogy.

In 1882, Edison lit the first electric lamps for customers in Manhattan. But in the thirty years after electrification, productivity hardly increased. The reason wasn’t that electricity was useless. It was that factory designs at the time revolved around steam engines. Only after factories were rebuilt did productivity improvements become apparent.

John’s judgment is that AI is at a similar stage. The transformation has already happened, but old models haven’t had time to absorb it. “However,” he said, “I think AI doesn’t need thirty years.”

Stripe’s data seems to confirm his optimism. The AI economy has already exploded on its platform. Nearly all traditional enterprises I’ve interacted with have top executives pushing AI deployment with a strong sense of urgency.

Born Global

Besides speed, another characteristic of these AI companies that left a deep impression on me is that they have been global from day one. Stripe calls this: go global by default.

Since becoming an AI blogger, I often have a certain experience: AI content creation is without time zones. News from across the Pacific is just as important as local news.

AI products operate similarly. Large language models blur the interface language and interaction habits that traditional software relied on. A unified chat window allows users worldwide to use products via natural language. In this sense, large language models have made a unified global software market possible for the first time.

Data from the conference confirmed this observation. In early waves, the fastest-growing SaaS companies covered about 25 countries in their first year, reaching 50 by year three.

AI companies develop at a completely different speed: reaching 42 countries in the first year, expanding to 120 by year three.

Maia said Kazakhstan is now appearing on many AI companies’ market lists. In the “Index Economy” session on the second day, Stripe provided a median figure: the top 100 AI startups sold products to 55 countries in their first year.

One company gave a concrete example. Emergent Labs was founded in the US in 2024, but nearly 70% of its revenue already comes from overseas. At least 16 countries contribute at least 1% of its income each.

Among leading AI companies, 48% of revenue comes from outside their domestic markets. Three years ago, this figure was only 33%. Global revenue is no longer supplementary; it’s the core.

Speed and globalization are two core features of the AI economy, both directly linked to Stripe. AI companies need to quickly establish payment capabilities. They need to accept payments in 40 countries and regions within their first week of founding. This is exactly what Stripe has been doing since its inception.

Here, let’s briefly revisit the background of Stripe’s founding.

Stripe’s founders, Patrick and John Collison, are Irish. They are inherently transnational entrepreneurs.

At the conference, I met an Irish peer who told me that in Ireland, these two brothers are considered heroes among AI entrepreneurs. When they moved to the US, they found online payment integration to be ridiculously difficult. Connecting to payment systems required bank contracts, PCI compliance checks, and multiple intermediaries—taking weeks or even months.

So, in 2010, two twenty-year-olds dropped out of college, moved to San Francisco, and wrote a solution with just seven lines of code that allowed developers to accept payments easily.

This seven-line code was born at the dawn of mobile internet and SaaS takeoff. Shopify needed to help millions of merchants collect payments. Uber needed frictionless payments for riders. Salesforce needed to handle global subscriptions.

They all chose Stripe. As Stripe grew alongside these global customers, it built local capabilities in 46 countries, covering 195 markets, and supporting 125 local payment methods.

For consumers, Stripe isn’t a household name.

It’s hidden behind Shopify’s checkout pages, OpenAI’s subscription confirmation emails, and Uber’s fare notifications. But this invisibility hasn’t prevented it from becoming the underlying financial pipeline of the internet economy.

In the AI era, this global financial infrastructure has given Stripe an early advantage in serving AI companies going overseas.

At this year’s conference, I also met Stripe’s Global Product Lead, Abhi Tiwari.

He took over this role three months ago and moved to Singapore. Stripe has engineering centers in San Francisco, Dublin, and Singapore, and a Latin American office in São Paulo. Abhi told me many AI companies start their conversations with Stripe the same way: “We are born global by default. It doesn’t matter where our users are.”

The old model of developing products at headquarters and then pushing them globally is being replaced by local teams directly developing in target markets.

Reaching global users is one thing. Collecting payments from them is another. The latter is quite complex because each market has its own currencies and payment habits.

Here, Stripe mainly helps AI companies and many other clients through two methods: pricing in local currencies and integrating local payment methods.

The former allows Brazilian users to see prices in BRL instead of USD, increasing cross-border revenue by 18%. The latter enables Indian users to pay via UPI, and Brazilian users via Pix, boosting conversion rates by over 7%.

After AI demo tool Gamma added UPI payments in India, its revenue in India surged by 22% that month. At the booth, I also saw Chinese company MiniMax. From what I understand, many Chinese companies going overseas use Stripe’s financial services through their overseas entities.

These AI-native companies also share another trait: very small teams. Many are solo founders. One or two people plus a group of Agents can run a truly global company with real revenue.

On the second day, Emily presented a data point: on Atlas, the density of solo founders is approaching 5,000 per million Americans, and more and more of them are earning over $100k annually.

She used the term solopreneur: a one-person company. John explained this phenomenon using Ronald Coase’s “Theory of the Firm”: firms exist because internal coordination costs are lower than market transaction costs.

But AI might overturn this logic. When Agents can discover services, integrate software, and handle payments for you, external coordination costs plummet. You no longer need a whole department of employees to do what once required a full team.

From Human Economy to Agent Economy

The AI economy described above, no matter how fast it develops or how globalized it becomes, still has humans as the primary actors. Humans buy AI products. Humans use AI tools to start businesses.

But at this year’s Sessions, the strongest signal I felt was that Stripe’s next major focus is another shift: an economic form where Agents become market participants. This is Agent Commerce.

This shift is already subtly reflected in Stripe’s own data.

Stripe Product and Business President Will Gaybrick showed a set of data. For years, Stripe’s command-line interface (CLI) has been used by a small group of tech-savvy users, with usage nearly flat.

But after 2026, usage suddenly skyrocketed. The reason is that Agents don’t need a fancy graphical interface. A simple CLI is often more practical.

Maia’s data shows that in 2025, the traffic of Agents reading Stripe documentation increased about tenfold.

If this trend continues, by the end of this year, the number of Agents reading Stripe docs will surpass that of humans. The API documentation Stripe spent over a decade refining is now finding a new, loyal audience.

If the idea of Agents spending money still feels unfamiliar, consider two real scenarios happening now.

The first is that shopping interfaces may have already shifted into model chat windows. Consumers now often search for products using ChatGPT, Gemini, or Instagram. The gap between search and transaction is being compressed into a single interface. China has similar examples, including the well-known story of buying milk tea within AI apps.

In a group interview, John Collison explained why this compression is hard to reverse, using his own experience of buying a travel power adapter.

If an Agent completes the entire process from search to order, and the product arrives at his home days later, he won’t go to another website to fill out personal info from scratch—even if that site’s product might be slightly better. Once the shopping Agent completes the search, the logical next step is checkout.

The second example is even more interesting: OpenClaw. Those who follow the “Claw” wave know it’s one of the hottest open-source autonomous Agent frameworks.

Users give instructions to the Agent via messaging apps like Feishu, Telegram, and WhatsApp, and the Agent autonomously executes tasks.

The key point is that OpenClaw can burn hundreds or even thousands of dollars’ worth of tokens in a single day. It manages token consumption and usage itself. While human authorization is still needed in many cases, ultimately, the tokens consumed are by the Agent. And tokens can be directly converted into money.

From managing token consumption to directly spending money, there’s only one step. At this year’s conference, Stripe demonstrated crossing that line.

Agent Buying and Selling

On the main stage on the second day, a demonstration drew applause.

John Collison gave a simple instruction to an Agent: research how AI demand affects the energy market. The Agent started searching, found an energy market data set from Alpha Vantage costing 4 cents, and autonomously purchased and downloaded it using a stablecoin wallet in Tempo CLI, since paying with a credit card for 4 cents is obviously unreasonable.

Then it generated a complete analysis report. Just this step was already astonishing. But John then told the Agent: “Publish and sell this report. Set a price you think is reasonable, and let other Agents find and buy it.”

The Agent checked the licensing terms of the Alpha Vantage data set, confirmed it was allowed for commercialization, built a website, published the report, and generated a command file that allows other Agents to purchase the data with a single request.

Within minutes, one Agent completed the entire chain: research, procurement, production, compliance, publishing, pricing, and sales. It was both buyer and seller. After the demo, John said: “Agent commerce is here.”

Two other demos on the first day were equally impressive. Will Gaybrick built an API code review app that allows Agents to perform review services for users. Throughout, he didn’t tell the Agent anything about payments.

During execution, the Agent automatically discovered that the app used a Machine Payment Protocol (MPP) and independently completed a $2 payment. The only human action was a fingerprint authorization. This zero-configuration payment discovery capability is core to the MPP protocol. Developers don’t need to write separate payment logic for Agents—they just find it.

Next, Gaybrick combined a real-time billing engine, Metronome, a blockchain designed for payments called Tempo, and stablecoins to demonstrate streaming payments—breaking down funds into tiny amounts that are transferred in real time and continuously as services like AI compute are consumed.

An app that charges based on AI token consumption at $3 per million tokens, with multiple Agents running simultaneously. The dashboard on the left shows token consumption rising steadily, while micro-payments in stablecoins flow in on the right.

On opening the Tempo blockchain explorer, you see a total payment of $3.30 composed of thousands of micro-payments under a dollar each, each about one-thousandth of a dollar.

Credit cards can’t do this. ACH clearing can’t. UPI and Pix can’t either. Gaybrick announced on stage that this is the world’s first streaming payment system.

The Return of Micro-Payments and New Consumption Logic

Shopping in chat windows and OpenClaw are examples of Agents representing human consumption. But in a group interview, Collison made a broader prediction: Agents might also create entirely new demands.

He believes Agents could revive a business model that has long been discussed but never truly succeeded: micro-payments. Humans are not good at making extremely fine-grained consumption decisions. Spotify replaced per-song payments with a $9.99 monthly subscription because no one wants to weigh whether a song worth 15 cents each time they press play.

Agents, however, have no such cognitive burden. If this judgment is correct, a whole class of business models that failed due to human cognitive resistance could suddenly become feasible in front of Agents.

Maia also expressed a similar view in a one-on-one conversation. She said she recently spoke with dozens of AI founders, and pricing was the most frequently discussed topic when they talked about Agent commerce.

Every transaction involves two parties: buyer and seller. If the buyer becomes an Agent, what should merchants do?

In an interview, I asked Stripe’s product lead Jeff Weinstein: humans often say “the customer is king.” Merchants need to please consumers. So how do they please Agents?

Jeff’s answer was to imagine Agents as the best programmers you know. They want perfect information, structured formats, fast readability, and all the background needed for decision-making.

Humans like beautiful images and smooth animations. Agents prefer raw structured data, precise logistics info, and want to complete transactions with as few steps as possible.

In another conversation, Meta’s VP of Product Ginger Baker summarized this shift more radically: payments will shift from a “moment” to a “strategy.”

Human purchases are discrete.

You walk to the checkout, take out your wallet, swipe your card, and the transaction is done.

Agent consumption, on the other hand, is continuous.

You set rules, such as “spend no more than $50 on household items this week,” “always use this card first,” or “never auto-authorize transactions over $500.” Then, the Agent autonomously makes ongoing purchases within your authorized framework.

Computing Power as the New Cash

If Agents truly become a new type of consumer, they will also bring new risks. These risks are fundamentally different from traditional SaaS transaction risks and are entirely distinct from human consumer risks.

During the Sessions, I paid special attention to this topic and discussed it with several Stripe executives.

Stripe’s data and AI head, Emily Glassberg Sands, described three rapidly growing fraud patterns. The first is multi-account abuse: the same person repeatedly registers different accounts, each gaining free credits.

According to Stripe’s network data, one in six AI company registrations involves such abuse. The second is malicious spending during free trials. This is especially deadly for AI companies because each trial burns real inference costs.

She gave an example: for a partner company, the token cost per paying customer exceeds $500, because it takes 25 free trials to convert one customer, with 19 of those being fraudulent.

The third pattern she called “freeloading”: customers consume大量tokens and then refuse to pay at the end of the month. Emily also quoted a phrase: “Compute power is the new cash.” When SaaS is abused, marginal costs are nearly zero. But each inference call in AI costs real money. Stealing tokens is like stealing cash.

However, there’s a particularly tricky dilemma here. Many AI founders respond to abuse by simply shutting down free trials.

Emily said she asked everyone claiming to have “solved” this problem how they did it, and found their solution was just to turn off the free tier. But Jeff argued this would trigger another problem.

Agents are increasingly the main way to discover new services. If an Agent can’t try a service on its own, it will just jump to another link.

Emily added that if the prompts given to Agents are “join the waitlist” or “contact sales,” they will leave immediately. Shutting down self-service registration to prevent fraud might mean handing over the most important growth channel to competitors.

Stripe’s solution to this dilemma is its fraud prevention system, Radar. The logic of Radar is simple: every transaction on Stripe is a learning opportunity.

Transaction data from 5 million businesses feeds into a shared risk recognition network. If one company encounters a fraud pattern, all benefit. Last month, Radar intercepted over 3.3 million high-risk free trial signups among eight high-growth AI companies.

Jeff also proposed a counterintuitive view: Agent shopping might ultimately be safer than human shopping on the web. Trust verification in human web shopping relies on inference: how long the user stays on a site, whether click paths look normal, etc.

But Agent transactions can be programmatically verified. Stripe’s shared payment tokens are tokenized, so Agents never see raw credit card numbers. Users authorize via biometrics, and can set transaction limits, time windows, and merchant whitelists.

When trust mechanisms shift from inference to confirmation, the security baseline might actually improve.

Ecosystem, Protocols, and a Piece of History

By now, it should be clear that realizing Agent commerce depends on a well-functioning ecosystem. At the 2026 Stripe Sessions, I met someone from the food industry. He said his purpose was to understand whether Agent commerce could become a new opportunity for his company. That’s the seller’s perspective.

So, this can’t be achieved by Stripe alone. It needs an ecosystem.

Walking through the exhibition hall for two days, I saw many booths from companies across the financial industry chain.

Stripe has also launched or joined a series of protocols with upstream and downstream partners to connect different parts of the ecosystem: buyers and sellers, humans and machines, and machine-to-machine. The Machine Payment Protocol (MPP) allows Agents to discover and complete payments via HTTP.

The Agent commerce suite enables consumers to make purchases directly within AI apps from Google, Meta, OpenAI, and Microsoft. The Universal Commerce Protocol (UCP), initiated by Shopify and joined by Meta, Amazon, Salesforce, and Microsoft, is a cross-platform commercial protocol. Stripe is on the UCP steering committee.

A group of companies, both partners and competitors, agree to collaborate on a shared protocol because fragmentation would make seamless cross-platform Agent consumption difficult. That’s not good for anyone.

Regarding protocols, I saw a special Stripe partner at the exhibition: Visa. To me, Visa is essentially a protocol platform.

Seeing Visa immediately reminded me of a book I’ve long admired: “Decentralized: How to Create and Manage a New Kind of Business in the Electronic Age,” written by Visa’s founder, Dee Hock.

One theme of the book is how banks, currencies, and credit cards can be redefined in the digital era. Money no longer has to be coins and paper bills. It can also be data that is institutionally guaranteed, recorded on networks, and flows globally.

In the late 1960s, US banks issued the BankAmericard, which expanded nationwide. Large cross-state consumers flooded in, causing the old system to collapse. Hock realized the problem was organizational structure. Dozens of competing banks needed shared infrastructure, but no existing organizational form allowed them to cooperate while competing.

He used decentralization principles to make all banks equal members of a new organization, and US Bank abandoned exclusive control over the system. That organization was later renamed Visa.

So, two different companies from different eras are doing similar things. Is there some kind of inheritance?

With the help of any Agent, the answer is easy to find. Patrick Collison has publicly paid tribute to Hock. After Hock’s passing in 2022, Patrick called him “a severely underrated innovator,” and said Hock inspired him and his brother.

A more obvious sign of influence is a hiring decision: David Stearns, the author of the authoritative academic history of Visa, later joined Stripe.

And a detail that would make anyone familiar with payment history smile: at the conference, Georgios Konstantopoulos, CTO of Tempo blockchain, showed the list of verifiers. One of the names was Visa.

The organization founded by Hock, Visa, has now become a participant node in a blockchain network incubated by Stripe. Students build a new network, and the teacher becomes a node.

When Patrick traced Stripe’s origins at the conference opening, he said he started as a Lisp programmer. One core idea of Lisp is “code is data.”

He translated this idea into Stripe’s own language: “Stripe’s fundamental philosophy is that money is data. When we launched Stripe in 2011, this was not the mainstream view in the industry.”

Hock explored the essence of money from an organizational theory perspective: money is simply a “guarantee of value exchange.” The medium that carries it can be anything. Collison, from a programming language perspective, directly equates money with data: a form of data that can be programmed, called via APIs, and operated by Agents.

Both express the same idea in different languages. On the same stage, Ginger Baker said more plainly: “Money is just another form of digital content.”

If money is data, then data consumers naturally become money consumers.

Stripe’s Content Gene

By now, the story of the AI economy is nearing its end. But let’s take a slight detour. Stripe can almost be considered a peer of content creators.

This company is not only good at financial services. It’s also very good at content products. Its publishing arm, Stripe Press, has excellent taste. Many people know it because of its publication, The Charlie Munger Collection.

Its podcast, A Cheeky Pint, is also distinctive and widely listened to. CEOs like Sundar Pichai, Dario Amodei, and Marc Andreessen have been guests.

During the conference, I met Tammy Winter, senior editor at Stripe Press, and designer Pablo Delcan. Tammy joked, “Stripe is basically a publisher with a multi-billion-dollar company attached.”

Pablo Delcan shared his understanding of taste. He said taste is something that develops over time and needs to be accumulated. Regarding design trends, he believes that the new challenge is how to add complexity through details and precision without sacrificing simplicity and clarity.

When the topic turned to books, Tammy told me that within Stripe Press, a series of books for founders and builders is called the “Turpentine” series.

These books focus on operational knowledge, tools, techniques, maintenance, and practical content that helps work run smoothly. They are not abstract theories. They aim to help readers solve specific operational problems.

The name comes from a story about Picasso: when art critics gather, they talk about form, structure, and meaning; but when artists gather, they talk about where to buy cheap turpentine.

This series aims to be the cheap turpentine in the hands of founders. Think about it: for AI companies going overseas, Stripe’s financial services are another form of turpentine. You don’t need to worry about payments, compliance, or foreign exchange. You can focus on building your product.

This anecdote may seem unrelated to the main story, but it actually has a potential connection.

Stripe also publishes a magazine called Works in Progress, whose core theme is how economies grow. Its podcast interviews leading figures in the AI economy. Sessions itself is somewhat like an economics lecture.

On the second morning, Patrick Collison spent the entire session discussing economic data, Coase’s theory of the firm, and Solow’s paradox. I suspect that a financial services company’s deep interest in economics stems from understanding structural shifts in the economy—precisely the way it finds its next product opportunity.

As a podcast enthusiast, when I saw John Collison on the first day, my most pressing question wasn’t about finance. It was about podcasts. I asked him, after interviewing so many different people, whether there was a fundamental question running through all those conversations.

He thought for a moment and said he was truly interested in how these companies actually operate, what kind of competitive equilibrium they are in, and how they understand their own business.

Coincidentally, a small episode appeared at the end of the first day. The last fireside chat was originally scheduled to be Patrick interviewing Greg Brockman, co-founder of OpenAI. But shortly before going on stage, the guest was replaced by Sam Altman. Patrick explained that “AI is a rapidly evolving field.”

Suddenly, it turned into a celebration. The audience cheered.

The two have known each other for nearly 19 years. Altman was one of Stripe’s earliest angel investors, investing when the Collison brothers were under 20. That’s why Altman appeared very relaxed throughout the conversation.

Near the end, Patrick asked a personal question: why did Altman invest in two teenagers? Altman said he remembered they wanted to build products that solved problems they personally encountered, and he saw the opportunity to scale because many others needed the same thing.

I think his answers about podcasts and investments point to the same thing: seeking genuine needs, solving real problems.

In the conversation, Altman divided OpenAI’s transformation into three stages: from a research lab, to a product company, to a “Token factory” providing intelligence to the world. Each stage corresponds to a different mission.

Stripe is the same.

In 2010, the problem for two Irish teenagers was that online payments were too difficult. Over the years, they solved the same problem for 5 million users. By 2026, they’ve discovered a new problem: their enterprise clients may soon no longer be human.

One is doing podcasts, the other publishing. On stage, they discuss Coase’s theory and Solow’s paradox; in the exhibition hall, protocols and APIs are everywhere. Stripe is not only creating the AI economy; it’s also recording it.

At the conference, a crazy idea flashed through my mind: Stripe controls transaction data amounting to nearly 2% of global GDP. It can see where every dollar of AI revenue comes from, where it goes, and how fast it’s growing.

If Solow had such a heartbeat monitor back then, perhaps he wouldn’t have needed ten years to find the footprints of the computer in the statistics.

Maybe one day, Stripe could provide a model for the AI economy—not a large language model, but a Nobel-caliber economic model. Who says that’s impossible? Just a few years before Demis Hassabis, founder of DeepMind, won a Nobel Prize, who could have foreseen that moment?