Stripe just launched Link, a digital wallet built specifically for the era of autonomous AI agents that shop, book, and buy on your behalf. According to TechCrunch AI, the financial services giant unveiled the product at its annual conference this week, pitching it as the secure middle layer between your payment credentials and the AI assistants now making purchases for you.
What stands out here is the timing. The number of people running always-on AI agents has been climbing fast enough that Apple sold out of its base model Mac Minis, a popular box for hosting these workloads. Plenty of those users have hesitated to hand over raw card numbers to a bot. Link is Stripe’s answer.
What Link Actually Does
On the surface, it looks like a standard digital wallet. TechCrunch AI reports it’s available on web, iOS, and Android, with the usual feature set you’d expect:
- Connect cards, banks, crypto wallets, and buy now/pay later services
- Store billing and shipping details for one-click checkout
- Track spending across accounts
- Manage recurring subscriptions, including auto-updating payment methods when cards change
- 90 days of purchase protection on eligible transactions from select merchants
The wallet handles the boring stuff well. But the agent integration is what makes it different from PayPal, Apple Pay, or any other consumer wallet on the market.
How the Agent Piece Works
Users grant an AI agent access to their Link wallet through an OAuth flow, the same standard auth pattern you’d use to connect any third-party app. From there, the agent can build a spend request and pass it back to you with context attached. You review, you approve, the payment goes through. Your card numbers never touch the agent.
Link works with autonomous agents like OpenClaw and others, per TechCrunch AI. Currently it supports traditional payment methods, with Stripe promising support for agentic tokens, stablecoins, and additional payment types “soon.”
Approvals come through as mobile or web notifications. You see the transaction details before any payment credential is shared. Stripe says it’ll eventually let users set custom spending limits or pre-authorize agents to act without approval inside defined boundaries.
The Infrastructure Underneath
Link runs on Stripe’s new Issuing for agents product, which spins up virtual cards that agents use for autonomous purchases. Each comes with real-time authorization, spending controls, and full transaction visibility. Two delivery options exist:
- Programmatic access to Link, which generates a one-time-use card per transaction
- Shared Payment Token (SPT), backed by payment cards and banks, for repeated use
This matters for developers too. Anyone building an AI personal assistant or agent product can plug into Link instead of building their own wallet from scratch. That’s a real shortcut. Building payment infrastructure is expensive, slow, and full of compliance landmines.
Why This Matters
The agent economy has a chicken-and-egg problem. Agents can’t be useful for commerce until they can pay for things. Users won’t let them pay until the security model feels right. Stripe is trying to solve both sides at once: a consumer wallet that handles agent permissions natively, plus the issuing infrastructure to let any developer ship agent-powered purchases.
This is significant because Stripe processes a huge chunk of internet commerce already. If Link becomes the default way agents transact, Stripe sits at the center of agent-driven shopping the way it sits at the center of normal e-commerce checkout today.
The Caveats
A few things to flag. Stablecoin and agentic token support isn’t live yet, just promised. The user-defined spending limits and pre-authorization controls are also a future feature, not a shipping one. Right now every agent purchase still needs manual approval, which limits how autonomous “autonomous” actually is.
And as TechCrunch AI notes, plenty of users still feel uneasy giving any wallet, agent-friendly or not, broad permission to spend. The OAuth model and one-time-use cards address the credential exposure problem, but they don’t address the trust problem.
More details at the original TechCrunch AI report.