Stripe has expanded its Link digital wallet into what it calls an AI-native payments layer, allowing autonomous agents to initiate purchases on behalf of users. The system enables agents to handle tasks like booking, subscriptions, and shopping while keeping payment credentials abstracted from the AI layer.
This is not a new wallet launch but a structural upgrade to Link, which already supports cards, bank accounts, BNPL, and checkout autofill across platforms.
The architecture is designed around controlled execution rather than full automation. Users grant access to agents via OAuth, after which the agent generates a spend request with context (merchant, amount, purpose). The transaction only proceeds after explicit approval.
Instead of exposing raw payment data, Stripe issues one-time-use virtual cards or shared payment tokens, ensuring that sensitive financial credentials never leave the wallet layer.
Stripe is entering this space with a distribution advantage. Link already serves over 250 million users globally, giving immediate scale to agent-based payments without requiring new user acquisition.
At the infrastructure level, Stripe processed $1.9 trillion in payments in 2025, highlighting the volume base on which this AI commerce layer is being deployed.
The strategic shift is from checkout optimization to programmable commerce. Stripe’s agent wallet is built on “Issuing for agents,” enabling developers to define rules such as spending limits, merchant restrictions, and real-time authorization controls at the transaction level.
This effectively turns payments into an API-driven workflow where AI systems can request, negotiate, and complete transactions within defined boundaries rather than relying on manual checkout steps.
Stripe is not operating in isolation. The company has already integrated agent-based checkout into ecosystems including Google’s AI products, alongside partnerships with platforms like ChatGPT and Copilot.
This positions Link as a backend layer for a broader “agentic commerce” stack where discovery, decision-making, and payment execution are handled within AI environments instead of traditional websites.
Despite the push toward automation, Stripe is maintaining a human-in-the-loop model. Every transaction currently requires approval, with plans to introduce conditional autonomy such as spending thresholds and predefined rules.
This reflects a constraint in the current market: while AI agents can technically complete purchases end-to-end, trust, fraud control, and liability frameworks still require user oversight.
Stripe’s Link upgrade signals a transition from user-initiated checkout to agent-mediated transactions, where AI systems act as intermediaries between intent and payment. The combination of tokenized payments, approval workflows, and large-scale distribution suggests that agentic commerce is moving from experimental to deployable infrastructure.
Discussion