Author: Emily Sands; Compiler: Peggy, BlockBeats

When AI Agents no longer just answer questions, but begin to place orders, deploy applications, call services, and consume tokens for people, the underlying rules of Internet commerce will also change.

This article comes from Stripe's redefinition of "AI economic infrastructure": in the past, payment infrastructure mainly served human users and software companies; now, it must also serve machine subjects. Agents need to be able to understand prices, complete payments, call wallets, and deploy services; AI companies also need to establish new billing, risk control, and settlement mechanisms around token consumption.

The Machine Payments Protocol, Link, Stripe Projects, Metronome, Tempo and streaming payments mentioned in the article essentially all point to the same trend: AI Agent is changing from a "tool" to a new economic participant in the Internet. They are both buyers and builders; they create software and consume resources. As a result, the traditional SaaS era business model of charging by seat, post-invoicing, and manual checkout is being replaced by a new model of real-time measurement, real-time settlement, and machine readability.

This is not only Stripe’s product narrative, but also a larger industry judgment: If a large number of software consumption in the future is initiated by Agent, then the payment, billing, wallet, risk control and deployment processes will all need to be redesigned. AI changes not only productivity, but also the way business itself is organized.

The following is the original text:

Agents are becoming buyers and builders. This will revolutionize the way business is done.

Stripe has been building economic infrastructure for AI over the past few years. This includes several levels: We help the world's fastest-growing AI company accelerate development through payment, billing, checkout, fraud prevention and tax infrastructure; we deploy AI into every aspect of the payment process to help companies increase profits; and we also enable Stripe to be natively embedded into the AI ​​tools used by developers.

But in the past six months, the meaning of the concept "economic infrastructure for AI" has changed. Intelligent agents are becoming a new agent on the Internet. This also brings us two new tasks: one is to help the agent purchase and build on behalf of individuals and enterprises; the other is to help enterprises adapt to the new economic model brought about by token consumption. In other words, Stripe is empowering agents to become independent actors; at the same time, it is also helping companies monetize products that are created and consumed by agents and are highly dependent on tokens.

Agent as buyer

E-commerce was originally designed for humans. Humans browse the website, click on price pages, enter card information, and complete the checkout process. But agents don't do this. What they need is a programmatic way to understand that a service requires payment, how much it costs, and how to complete the payment without a human clicking through the checkout screen.

That’s why we partnered with Tempo to create the Machine Payments Protocol. It allows businesses to programmatically accept payments directly from agents without human intervention. There are no account registrations, no checkout pages, and no humans in the process; just a machine-readable payment method that allows agents to purchase services from businesses.

Once businesses can accept payments from agents, an equally important question is: How can consumers securely authorize agents to spend money on their behalf? This is where Link, our consumer wallet, comes into play.

Today, more than 250 million people are using Link. We are enabling Link to also serve agents, allowing users to authorize agents to securely complete payments on their behalf. Humans remain in control, while agents gain the ability to spend on behalf of the user.

Helping agents complete purchases on the Internet is a way for us to empower agents. Another way is to help them build.

Agent as builder

Today, vibe-coding is easy, but vibe-deploying is not. Agents can generate an application in minutes, but actually deploying it to the Internet still requires a lot of manual connections: creating accounts, configuring services, managing credentials, accessing APIs, and switching back and forth between different consoles.

Stripe Projects allows developers and their agents to register, manage and integrate various services required to deploy applications directly through the command line. The goal is simple: make vibe-deployment as easy as vibe-coding.

Token Realization

As agents begin to become buyers and builders, companies also need to adapt to a new world where the inference cost, or token cost, of a product changes dynamically. This will change the economic model of software.

In the SaaS model, serving one more user usually brings little additional cost. But in AI products, there is a real marginal cost behind every prompt word input, every API call, and every agent task. Therefore, enterprises can no longer just bill by seat, but need to measure usage in real time and charge customers based on the resources they actually consume.

This is why usage-based billing has become the core of AI monetization. Businesses can charge around the value that customers truly value: usage, workflow, results, or whatever unit of billing best suits their product.

Lovable is a good example. It started with a simple subscription model on Stripe to enable quick monetization after launch. As the company grew, so did its billing model. Now, when customers exceed the usage limit included in the subscription plan, Lovable charges based on the AI ​​tokens consumed. ElevenLabs has followed a similar path: it initially adopted a subscription model on Stripe, then introduced pay-as-you-go pricing as the product shape and customer usage patterns changed.

Token theft

But the premise of usage-based billing is that the company can really get the money back. Today, fraudsters no longer just steal funds or account credentials; increasingly, they steal tokens. As long as they can register an account, quickly consume a large amount of tokens, and then disappear before the bill is due, the economic model of AI products will quickly collapse.

Token theft is one of the most under-discussed issues in the AI industry today. Stripe Radar, our fraud protection product, evaluates new accounts in real time, predicts which free trials are potentially abusive, and identifies non-payment risks as usage accrues.

Streaming payment

The abuse problem becomes even more problematic when the clients themselves are agents. Agents can consume tokens at machine speeds, making traditional billing methods too risky or even unaffordable.

Businesses can require users to pay in advance and stop services when the limit is reached. Doing so protects the business, but creates a worse customer experience and prevents good customers from increasing their spending. Enterprises can also allow usage to continue to accumulate and then issue unified invoices. This is a better experience for the customer, but the bill may ultimately never be recovered, and by that time, the tokens will have long been consumed.

The better answer is actually obvious, but it is technically very complex for companies to implement it themselves: tracking usage in real time and collecting payments in real time. This is exactly what the combination of Metronome and Tempo enables.

Metronome is a company acquired by Stripe that mainly serves the most complex usage billing model, which can track usage in real time as tokens are consumed. Tempo supports low-cost, high-frequency stablecoin payments and can achieve instant settlement. After the two are combined, the AI ​​company can complete the charging while the token is consumed, without having to choose between "hard cap" and "non-recoverable bill afterwards".

We call it streaming payment: a new business model for AI-native business and adapted to machine-speed software consumption.

The above is exactly what today’s “AI economic infrastructure” needs to carry.

It no longer just provides payment services for AI companies, but provides a business system for agents, provides wallets for agents, allows agents to deploy software, supports billing around tokens, prevents fraud and abuse in token use, and provides streaming payment capabilities for agents.

AI is changing business and the way businesses are built. Therefore, the infrastructure must also change.

Stripe is building it all.