← Articles

Stripe Link and AI Agents: What a Wallet Supplies and What Agent Purchases Need

Published July 29, 2026Updated July 29, 2026Analysis

Key takeaways

  • Stripe Link is a consumer payment wallet. Since April 2026, its agent wallet lets an agent request a one-time-use payment credential for a specific amount and seller, after the user approves the request.
  • An AI agent purchase needs more than the payment step. At many merchants, something must also sign in, build the cart, complete the checkout form, and get past bot detection.
  • The useful distinction is payment capture versus purchase execution. A wallet supplies the credential for the payment. An agent must also execute the checkout.
  • This is a question of scope, not quality. Link solves the credential and approval problem well, and agent purchases add requirements that sit outside the payment step.
  • Stripe also takes part in agentic commerce through the Agentic Commerce Protocol, which it co-developed with OpenAI.

Stripe Link gives an AI agent an approved payment credential, but at most merchants the agent must still reach the payment step on its own. The agent creates a spend request, the user approves it, and Link returns a one-time-use credential. Before payment, the agent must still sign in at the merchant where an account is needed, build the cart, complete the checkout flow, and get past bot detection.

This page compares scope, not quality. Consumer wallets such as Link solve a real problem well. Agent purchases add a different set of problems, and it helps to see where each one starts.

Stripe's products change over time. This page describes Link and its agent wallet at a general level, as of November 2026. For current features, read Stripe's own documentation.

What is Stripe Link designed to do?

Stripe Link is designed to let a shopper save payment and shipping details once and reuse them to pay quickly, and its agent wallet lets an agent pay with those details after the user approves. Both jobs center on the payment step.

  • Saved details across merchants: A shopper saves a card or other payment method once, and the details are available at the next checkout that offers Link.
  • Fast pay for returning shoppers: The flow assumes a person who is on the merchant's checkout page and ready to pay.
  • Merchant-side adoption: The merchant enables Link, and it appears as a payment option inside the merchant's existing checkout.
  • A person present at sign-in: Verification, such as a one-time code to the shopper's phone or email, assumes the shopper is there to receive and enter it.
  • An agent wallet: An agent creates a spend request with an amount, a seller, and a description. The user approves it on the Link website or app, and Link returns a one-time-use virtual card or a token. The user's real card number is not exposed.

Each of these properties is a sound choice. Together they describe a product that helps at the moment someone, a person or an agent, reaches the payment step.

What does an AI agent need before payment?

Before payment, an AI agent needs to sign in to the merchant, build the cart, complete the checkout flow, identify itself, and stay within what the user approved. A person does all of this before they use a wallet. For an agent, none of it happens automatically.

  1. 1

    Sign in to the merchant: Real prices, saved addresses, delivery windows, and loyalty terms often sit behind an account session. Sign-in can also need a one-time passcode sent to the user.

  2. 2

    Build the cart: The request must resolve to specific products, variants, quantities, and fulfillment options at that merchant. A payment method does not choose a size or a delivery slot.

  3. 3

    Complete the checkout flow: Checkout has several steps: address validation, shipping or delivery choice, tips, taxes, and confirmation. Each merchant builds these steps differently.

  4. 4

    Identify the agent: Merchants and their bot detection systems treat unidentified automated traffic as a risk. An agent that cannot prove who it is may be blocked before it reaches payment.

  5. 5

    Stay within what the user approved: The purchase must stay within what the user authorized: this merchant, these items, this amount. A one-time credential can cap the amount and name the seller, but the agent must still put the approved items in the cart.

A wallet supplies the payment credential. An agent must complete every step before the payment.

What is the difference between payment capture and purchase execution?

Payment capture collects and charges a payment method once a purchase is ready to pay for, and purchase execution produces that purchase in the first place. Execution covers sign-in, the order, the merchant's controls, and the confirmed transaction under approved terms.

RequirementConsumer wallet such as LinkAgent purchase execution
Store payment details for reuseYes, a core purposeNeeded, with the card tokenized so the agent never sees the number
Give the agent a credential after user approvalYes, with the agent walletNeeded
Reduce friction for a returning personYes, a core purposeNot the main goal; the user approves terms and the agent acts
Sign in to the merchant for the userOutside its scopeNeeded
Add items to a cart at the merchantAt merchants that support UCP; elsewhere the agent does thisNeeded
Complete a multi-step checkoutAt Stripe checkouts and UCP merchants; elsewhere the agent fills the formNeeded
Present a verifiable agent identityBeta attestations that a service can verifyNeeded, for example with signed requests
Record what the user approvedYes, an amount, a seller, and a descriptionNeeded, for example with an approved line-item intent

Read the table as scope, not as a score. A payment product has no reason to sign in to merchants or build carts at every site. An agent removes the person from those steps, and that creates new infrastructure requirements rather than a need for a new payment method.

Is Stripe working on agentic commerce?

Yes. Stripe offers the Link agent wallet, and it co-developed the Agentic Commerce Protocol (ACP), an open specification, with OpenAI. ACP is a merchant-side protocol: it defines how an agent reads a merchant's products and completes a checkout through structured endpoints, at merchants that implement it.

A merchant-side protocol and a wallet solve different problems. A protocol helps at merchants that adopt it. An agent that must also buy from merchants without such an endpoint still needs sign-in, checkout, identity, and consent handled in another way.

What does an agent purchase system need in addition to a wallet?

An agent purchase system needs a verifiable agent identity, a recorded approval of specific terms, a payment credential the agent cannot read, and a way to complete the checkout at the merchant.

  • Verifiable agent identity: The agent signs its requests with a published key, for example with HTTP Message Signatures (RFC 9421) as used by Web Bot Auth, so a merchant or CDN can check who sent them.
  • An approved intent: The user's request becomes a structured record of merchant, items, and amounts. The user approves it before the purchase, and the system checks execution against it.
  • A credential the agent never sees: The user's card is tokenized in a PCI DSS compliant vault, so the agent and the agent builder never handle the card number.
  • Checkout completion: Through a merchant protocol where one exists, or through the merchant's own checkout where it does not, the system reaches a confirmed order.
  • A clear approval moment: The user confirms the specific terms in the agent's own interface, often with a passkey or biometric.

When the charge runs on the user's own card, the user keeps the normal dispute rights of that card. That matters, because users are asked to let software spend their money, and the answer to "what if something goes wrong" should be the familiar one.

How should a builder decide which one they need?

A builder should decide by who drives the purchase. If a person is on your checkout page and you want them to pay faster, a consumer wallet such as Link fits the job. If your agent needs an approved credential, the Link agent wallet supplies one. If software completes the purchase for a user at a third-party merchant, you need purchase execution as well.

One test helps: ask which party is responsible for reaching the merchant's order confirmation. If the answer is the user, you have a payment capture problem. If the answer is your agent, you have an execution problem.

The two are not exclusive. A merchant can offer a wallet to human shoppers and a structured checkout to agents at the same time.

Frequently asked questions

Can I use Stripe Link to let my AI agent buy things?

Stripe Link supplies the payment part. With the Link agent wallet, the user approves a spend request, and Link gives the agent a one-time-use credential. At most merchants, the agent must still sign in, build the cart, complete the checkout form, and get past bot detection.

Is Stripe Link a poor fit for agentic commerce?

Stripe Link is not a poor fit. It does its intended job well: it reduces friction in the payment step, for human shoppers and, through its agent wallet, for agents. Agent purchases add requirements outside that step, such as merchant sign-in, cart building, and checkout completion.

What is the difference between payment capture and agent purchase execution?

Payment capture collects and charges a payment method once a purchase is ready to pay for. Purchase execution produces the purchase: sign-in at the merchant, the order, the merchant's controls, and a confirmed transaction under terms the user approved.

How does Stripe take part in agentic commerce?

Stripe takes part in agentic commerce through the Link agent wallet, which gives agents approved payment credentials, and through the Agentic Commerce Protocol (ACP), an open merchant-side specification that Stripe co-developed with OpenAI. ACP defines how an agent reads a merchant's products and completes checkout at merchants that implement it.

Why does agent identity matter if the payment credential is valid?

Agent identity matters because merchants can block an agent before payment, whatever credential it holds. Bot detection, one-time passcodes, and CAPTCHAs stop unidentified automated traffic. A signed, verifiable agent identity gives the merchant something to check.

Related articles