Stripe’s Link now works for AI agents — and that’s a big deal

Stripe just made a move that feels both inevitable and overdue: its Link digital wallet now supports autonomous AI agents.

For anyone who hasn’t been following Link closely, it’s Stripe’s take on a universal checkout wallet. You store your cards, bank accounts, and subscription info once, then use it across any merchant that accepts Stripe. No more typing in your card number for the hundredth time. It’s been around for a while, mostly for human shoppers.

Now the same infrastructure is open to AI agents. The idea is straightforward: an agent can initiate a payment via Link, but the actual spend requires human approval through a configurable flow. You set the rules — spending limits, merchant whitelists, maybe a per-transaction approval — and the agent operates within those guardrails.

This is higher than I expected from Stripe, honestly. Not because they lack capability, but because most payment companies have been tiptoeing around the agent use case. They talk about APIs and machine learning but rarely ship something this concrete for autonomous spending. Stripe just did.

The approval flow is the key piece. Without it, this would be a non-starter for anyone sane. You can’t have an AI agent with unfettered access to your bank account. But a system where the agent requests, you approve (or set standing rules), and the transaction goes through? That’s workable. It’s basically a programmable version of how you’d handle a corporate credit card for a junior employee.

I’ve seen this approach tried before in the crypto space — smart contract wallets with spending limits and guardian approvals. But those never hit mainstream because the UX was terrible and nobody actually uses crypto for daily payments. Stripe has the advantage of existing merchant adoption and real payment rails. Link already works with millions of businesses. Adding agent support means those businesses don’t need to do anything new.

There’s a subtle shift here that matters. Most “AI commerce” announcements so far have been about agents finding products or comparing prices. That’s the easy part. The hard part is the actual transaction — the authorization, the fraud checks, the settlement. Stripe is solving that part, which is where the real bottleneck was.

Will this lead to a flood of autonomous shopping bots? Probably not immediately. The approval flows are still going to annoy people if agents are constantly requesting permission for small purchases. But for recurring subscriptions, restocking supplies, or managing cloud costs? That’s where this makes sense. Let your agent handle the predictable stuff, approve once, move on.

Stripe isn’t the only player here, but they’re the first at scale. PayPal has similar ambitions with their wallet products, and you can bet Adyen and others are watching closely. The difference is Stripe’s developer ecosystem. If you’re already building on Stripe, this is a one-line integration. That matters more than any feature comparison.

I do wonder about the edge cases. What happens when an agent tries to spend at a merchant that’s fraudulent? Stripe’s fraud detection should catch some of it, but agents can be manipulated in ways humans aren’t. Prompt injection attacks that trick an agent into buying something it shouldn’t. Stripe will need to stay ahead of that.

For now, this is a solid step. It’s not flashy, not trying to reinvent money or launch a token. Just practical infrastructure for a world where agents will need to pay for things. That’s exactly what the space needs right now.