Stripe's Link Wallet Now Lets AI Agents Spend on Your Behalf

Autonomous AI agents can now book, buy, and pay—and Stripe just built the financial infrastructure to make that less terrifying than it sounds.

At its annual Sessions conference, Stripe announced a significant upgrade to Link, its existing digital wallet. The new version lets users connect AI agents—like OpenClaw and others—and authorize them to make purchases without ever exposing raw payment credentials. Shopping, reservations, tickets, subscriptions: if an agent can initiate the transaction, Link can now handle the payment layer with user approval built in.

How the Credential Problem Gets Solved

The mechanism matters here. Users grant their agent access to Link via a standard OAuth flow. The agent creates a spend request, surfaces the context to the user, and waits for approval before any payment credential is shared. On mobile and web, that approval comes as a notification—review, confirm, done.

Under the hood, Link runs on Stripe's new Issuing for Agents infrastructure, which issues virtual cards for agent use with real-time authorization, spending controls, and full transaction visibility. Developers can give agents programmatic access to Link for one-time-use cards, or use a Shared Payment Token backed by payment cards and banks. The agent never holds your actual credentials. It holds a permission that expires.

Stripe says future controls will let users set spending limits and define conditions under which agents can act without per-transaction approval.

Why This Infrastructure Moment Matters

The timing is not incidental. Apple recently sold out of its base-model Mac Minis—a popular platform for running always-on AI agents—which signals that adoption of autonomous agents among consumers and professionals is moving from early-adopter territory into something broader. The bottleneck was never the agent's ability to act. It was the trust layer required to let it act on anything financially consequential.

Stripe is building that trust layer. An agent that can research, decide, and execute—but stops at the payment step for human approval—is a meaningfully different proposition than one that requires you to hand over a card number and hope for the best. The approval-first model preserves human oversight at the moment of financial commitment, which is precisely where most people draw the line.

What This Means for Marketing and Growth Teams

The near-term marketing applications are narrower than the headline suggests—most campaign spend still runs through platforms with their own authorization flows. But the medium-term implications are real.

As AI agents become standard tools for tasks such as media-buying research, vendor outreach, event registration, tool procurement, and subscription management, the payment authorization layer becomes a workflow design question, not just a technical one. Who approves agent spend? At what threshold? Under what conditions does approval become automatic? Those are governance questions your team will need answers to before autonomous purchasing becomes routine.

Link also gives developers and businesses a ready-made wallet infrastructure rather than building their own—which lowers the barrier for anyone building an AI assistant with a commercial action layer. Expect more agent-native products to appear that assume payment capability as a default feature rather than an advanced add-on.

The agentic commerce era has a financial rail now. The strategic question is whether your organization has thought through what it wants agents to be authorized to do.

As AI agents move into commercial workflows, the teams with a clear strategy for what to automate—and what to keep human—will have a serious edge. Winsome Marketing's growth strategists can help you build that framework. Let's talk.