Stripe has extended Link, its 250-million-user consumer wallet, to AI agents. Starting at Stripe Sessions 2026, users can authorize an agent to make payments through Link on their behalf — with the wallet issuing a one-time-use card or a Shared Payment Token (SPT) per task, backed by the cards and bank accounts already inside the wallet. The consumer's real payment credentials never leave Stripe.
The security model is the headline. Each agent purchase is gated by a per-transaction approval and a single-use credential, so a compromised or hallucinating agent can't drain a card or persist credentials beyond the one task it was authorized for. That solves the trust problem that has been blocking serious commercial agent deployments — companies don't want their agents holding real card numbers, and consumers don't want autonomous software charging arbitrarily.
Integration is deliberately small. The @stripe/agent-toolkit (Python and TypeScript) plugs into LangChain, CrewAI, and similar frameworks via function calling, and Stripe shipped a SKILL markdown file alongside it so agent runtimes that support Anthropic's Skills format can adopt the wallet with minimal glue. The whole onboarding fits in a single npm install plus a SKILL file.
The strategic frame is that Stripe is positioning itself as the economic substrate of the agent era — alongside 287 other AI-related launches at Sessions 2026. The bet is that agents will need to transact constantly (subscriptions, API calls, micro-purchases, marketplace buys), and whoever owns the trust layer between human, agent, and merchant captures a disproportionate share of that flow.