Agents are a new kind of economic actor on the internet. For decades, businesses have built for two kinds of customers: consumers and other businesses. Now, agents are beginning to research, compare, provision services, and buy on behalf of both people and businesses. They can help people discover and buy products, create an account for a SaaS service, purchase a license and more.

This helps businesses reach new customers as they use the internet to get things done. More people are asking agents for help, then relying on them to find and buy products or set up the services they need. Businesses that can work with agents can serve these customers and make more sales.

But the internet was designed around human workflows: webpages, forms, logins, and checkout flows. While agents can crawl websites, the resulting data quality is sometimes flawed; agents prefer structured information, programmable interfaces, clear access and permissions, and ways to initiate transactions.

Stripe is building the infrastructure that makes this possible: helping businesses become discoverable to agents, letting agents provision accounts and access services, enabling secure and programmatic checkout via APIs, and helping businesses process payments with defined permissions and fraud protection.

This primer is a practical guide to that emerging stack: what capabilities exist today, how they fit together, and how different businesses can get started.

Emerging agentic modalities in the internet economy

Agentic commerce is emerging across three overlapping AI modalities: the browser, LLMs, and headless agents. Each represents a different way people and agents discover, evaluate, and buy from businesses. The browser remains essential to find information and check out, LLMs are becoming a new discovery surface, and more capable agents are beginning to take action on behalf of a human. These modalities will coexist, each playing a different role:

1. Browser: The standard model for online commerce today. Customers manually search for products, visit websites, and complete checkout flows themselves. Today, agents use browsers to scrape data and fill out forms as if they were human, rather than using programmatic, structured instructions.
   
2. Conversational LLMs: AI platforms such as ChatGPT and Gemini give customers a conversational way to describe what they need, research options, and receive recommendations. In this model, agents primarily read, reason, and summarize information to help a customer decide what to buy. Today, most LLMs then redirect a consumer to a business’s website to buy, putting more emphasis on having a sleek ecommerce checkout.
   
3. Headless: More advanced AI assistants, like Cursor or Claude, can read information and write, provision services, and transact on a consumer’s behalf. These interactions are often programmatic rather than browser-based, meaning an agent can discover a business, compare options, create an account, make a purchase, or pay for a service through structured interfaces and defined permissions. The human may set the intent, budget, and approval policy without manually completing every step of the transaction.
   

Sellers need to be ready to interact with agents through any of these modes. That means making products, pricing, availability, and purchase flows machine-readable and offering programmable ways to transact.

The four capabilities required for agentic commerce, and how Stripe helps with each

Agentic commerce depends on four building blocks:

• Discovery, so agents can find and understand a business
  
• Fraud protection, so businesses can identify suspicious activity
  
• Checkout, so agents can complete purchases while sellers retain control of the order
  
• Payments, so agents can pay without accessing a customer’s card or bank details.
  

The quickstart guides below explain how to apply each one to your use case.

1. Discovery

Agents can only recommend, provision, or transact with businesses they can find and understand. Discovery depends on accurate, structured information about what a business sells, how customers can access it, and how an agent can interact with it.

Stripe supports discovery across the emerging use cases:

• Stripe Directory is the primary starting point for agentic discovery. It is a universal registry of businesses that can transact with agents via an API endpoint. Agents can query it to find services, understand their capabilities, and identify available ways to transact.
  
• Catalog feeds help retailers and other businesses selling physical goods distribute structured product information to AI platforms. Participating merchants can share current product details, pricing, availability, and inventory with platforms that ingest these feeds, giving agents more accurate information than web scraping or unstructured content alone.
  
• Stripe Projects is designed for software providers whose services require account creation, provisioning, and credentials before use, and people and agents building their tech stack. Projects creates an account with a software provider on their behalf, retrieves the necessary credentials, and sets up billing programmatically. This is particularly relevant for services such as hosting, databases, authentication, analytics, and other tools an agent may need to assemble or operate an application. Stripe Projects providers also appear in Stripe Directory, making them discoverable alongside other machine-ready businesses.
  

2. Fraud

As agents take more consequential actions, such as browsing, creating accounts and making purchases, businesses need to know whether the activity is legitimate. The challenge is distinguishing authorized agent activity from abuse.

• Stripe Radar evaluates agentic transactions using the same fraud infrastructure that protects conventional payments. It assesses signals such as dispute likelihood, card testing patterns, stolen-card risk, and issuer-decline patterns, then applies your existing rules and custom logic to agentic orders. Shared Payment Tokens (SPTs) further limit risk by restricting a token to a specific seller, amount, and time window.
  

Agents also need a source of funds. Agentic wallets can hold or request funds and provide secure payment credentials without exposing a buyer’s underlying card or bank details to the agent.

• With Link's wallet for agents, buyers review and approve each spend request, while Stripe uses a secure payment token or one-time virtual card for the transaction. For businesses that want to embed a wallet in their own product and control its branding and user experience, Privy provides embedded wallet infrastructure with support for stablecoins and crypto networks.
  

Wallets work well for frequent or small purchases: owners can fund a wallet and set spending limits, enabling autonomous payments within those limits. For larger purchases, they can require approval for each request or set a task-specific budget.

3. Checkout

Traditional checkout was built for people filling out forms. Agent-ready checkout gives sellers defined, reliable ways for agents to complete a purchase while the seller retains control of pricing, inventory, fulfillment, tax, refunds, and the customer relationship.

There are three common checkout models, corresponding to the three internet models above. It’s important for businesses to understand all the different ways that agents may engage their brand, and ensure that they are ready to sell through any checkout:

• Browser checkout. A conversational LLM helps a consumer discover products—for example, Gemini recommends a book—and then sends them to the merchant’s website to complete the purchase. Businesses need a fast, reliable checkout experience on the business’s own site for converting this traffic. Stripe’s Optimized Checkout Suite helps reduce friction by showing relevant payment methods and offering accelerated checkout with Link, a digital wallet built by Stripe.
  
• Embedded checkout. An AI platform can keep the customer in its own interface, reducing handoffs, while using Delegated Checkout to request product, fulfillment, shipping, and payment options from the seller. The customer selects the relevant options and approves the order; the seller remains the merchant of record and continues to manage tax, fulfillment, refunds, and customer service. AI platforms use Shared Payment Tokens, a new agentic payment primitive, to initiate payments using a buyer’s saved payment method without exposing payment credentials. And businesses can use Stripe Tax to calculate the right amount and report on taxes.
  
• Headless checkout. A more autonomous agent, like Grok Bot or Claude, can transact through a programmatic interface rather than a browser or conventional user interface. For paid APIs and services, the business can return a 402 Payment Required response and accept payment through the Machine Payments Protocol (MPP) or x402. This model is designed for machine-speed purchases, including per-request, per-session, or other usage-based transactions.
  

4. Payments

How an agent pays depends on how the seller makes itself available—through conversational AI or headless agents. In conversational AI, agents either redirect to a business’s website or, increasingly, use an embedded checkout within the chat interface. Headless agents use APIs to process programmatic payments without a traditional checkout session. In both cases, agents can pay without exposing a customer’s underlying payment credentials.

• For payments conducted via an embedded checkout, like a checkout page within Gemini, Shared Payment Tokens (SPTs) let an agent pay with a customer’s preferred saved payment method without seeing their payment credentials. A consumer could ask an agent to find a carry-on suitcase that will arrive before an upcoming trip. The agent completes the purchase with an SPT limited to that retailer, order amount, and a short time window. The retailer still manages shipping, tax, returns, and support as usual.
  
• For payments conducted via headless checkout, a business can charge per request instead of requiring it to create an account or go through a traditional checkout. For example, an agent might ask a shipping provider’s API for a real-time quote. The API responds with a 402 Payment Required message—a prompt that tells an agent it needs to pay before it can fulfill a request— and specifies the price. It pays through the Machine Payments Protocol (MPP)—using stablecoin or an SPT—and gets the quote immediately.
  

Quickstart

These quickstart guides translate agentic commerce into practical starting points for different kinds of businesses, including becoming discoverable, enabling checkout, payments, and fraud controls.

Choose the guide that best matches your role:

• AI platforms: Help consumers discover and buy in your website or AI experience
  
• Software providers: Sell services to developers and agents
  
• Retailers and ecommerce businesses: Sell wherever customers discover products
  
• Businesses beyond retail and software: Prepare for agent-led discovery and transactions
  

And for more best practices about how to get started in agentic commerce, see How to get ready for agentic commerce in an evolving landscape.

Glossary

Agentic commerce: Stripe's solution that allows businesses to sell to and through agents, making their products discoverable, giving them controls to manage fraud and enabling agentic payments.

Discovery:

• Catalog feed: Structured product data that businesses connect to Stripe to distribute their products across agents in near real-time.
  
• Stripe Directory: A universal registry for machine-ready businesses that allows agents to discover and interact with them.
  
• Stripe Projects: A developer platform that lets software providers connect structured product data to Stripe, making their software services—hosting, databases, authentication, AI, and analytics—discoverable and available to people and their agents in near real time.
  

Fraud:

• Stripe Radar: Fraud-prevention tools that help detect bot abuse so businesses can distinguish between legitimate agentic payments and fraudulent bots.
  

Checkout:

• Optimized Checkout Suite: A suite of ecommerce tools that reduce friction and improve conversion for consumers referred to a business’s website by an agent, before or instead of embedded checkout.
  
• Delegated Checkout: An API that allows agents to embed secure checkout within their AI interfaces.
  
• Machine payments: Programmatic payments that let agents pay directly for products or services without browser checkout or human approval for every transaction. Using protocols such as the Machine Payments Protocol (MPP), they support machine-speed, usage-based purchases like per-request or per-session payments. They can be initiated and settled in fiat while using stablecoins for processing.
  
• Stripe Tax: A compliance solution that calculates tax in embedded checkout flows based on tax registrations, product details, customer location, and fulfillment information.
  

Payments and wallets:

• Shared Payment Tokens (SPTs): A payment primitive for agentic commerce that lets agents initiate payments with a customer's preferred payment method without exposing the underlying credentials. Scoped to a specific seller, amount, and time window.
  
• Link's wallet for agents: A machine-native way to let agents make purchases on behalf of a human without sharing the underlying payment credentials.