Agentic commerce arrives: Stripe’s new payment primitives

Stripe has been laying the groundwork for AI-driven purchasing with the Agentic Commerce Protocol (ACP), an open specification developed alongside OpenAI. The first production use case is now live: Instant Checkout in ChatGPT, letting US users buy directly from US Etsy sellers, with support for Shopify merchants like Glossier and SKIMS rolling out soon. But ACP is only the starting point for a broader set of tools designed for both AI platforms and the businesses selling through them.

Shared Payment Tokens: explicit trust for automated purchases

Traditional commerce assumes that possession of a credential is proof of intent. AI agents break that assumption—they act on behalf of users, so trust must be granted explicitly rather than inferred. Stripe’s answer is Shared Payment Tokens (SPTs), a payment primitive built for agent-initiated transactions. SPTs let an agent charge a buyer using their saved payment method without ever exposing the underlying card details.

SPTs are designed around five key properties:

  • Programmable: Tokens can scope payments to a specific business, cap usage by time or amount, be revoked on demand, and emit webhook events for monitoring.
  • Reusable: Buyers can use existing saved payment methods or add new ones without re-entering details for each purchase.
  • Secure and interoperable: Credentials are never exposed, and SPTs are built to work with card networks’ own agentic token standards.
  • Fraud-aware: On Stripe, SPTs integrate with Radar and carry risk signals—fraud dispute likelihood, card testing, stolen cards, issuer declines—to separate legitimate agents from low-trust bots.
  • Easy to adopt: SPTs plug into existing PaymentIntents flows, so developers don’t need to learn a new integration model.

The payment flow works like this: a buyer saves a payment method with the AI platform; when they approve a purchase, the platform issues an SPT scoped to the seller with a usage limit matching the checkout total; the SPT identifier is sent to the business via API; the business creates a PaymentIntent with it; and Stripe handles fraud detection and enforces token controls in real time.

Merchants that don’t process on Stripe can still forward SPTs to their own vaults or alternative processors as secure credentials, preserving the same programmatic trust controls and fraud protections.

Business controls: catalog, checkout, and customer identity

Early testing revealed three areas where businesses want more say in agentic commerce:

Product catalog structure

Stripe is designing product feeds that let merchants define how their goods appear across AI channels, including the ability to broadcast real-time pricing and availability without losing control of brand presentation.

Brand-preserving checkout

Rather than being intermediated by agents, businesses want to retain their brand, upsell opportunities, and customer relationships. Stripe is building an extensible checkout that embeds inside AI agents while letting merchants dictate its look, contents, and backend connections.

Identity and trust management

Stripe is extending the SPT foundation to Link, its digital wallet, turning it into a wallet purpose-built for agentic commerce. Customers would see their purchase history and identity persist across AI surfaces, with accelerated checkout and explicit controls like spending limits or per-agent permission grants. Stripe validates business identity on one side and confirms buyer authorization on the other.

Early partners and rollout

Stripe is already working with Microsoft Copilot, Anthropic, Perplexity, Vercel, Lovable, Replit, Bolt, Manus, and global payments networks to test agentic commerce in real-world scenarios.

Existing Stripe payment customers will be able to enable agentic payments in their current integration with as little as a one-line code change. Businesses using other payment providers can still adopt Stripe for agentic transactions. The full technical details are available in the official documentation.