Payment infrastructure for agents
Agentic commerce creates a trust problem of its own: how do you let a piece of software charge a
user’s card automatically, without confirming every single transaction? Visa and Mastercard answer
that at the payment infrastructure level.
Visa Intelligent Commerce and Mastercard Agent Pay are programmes that let issuing banks
create dedicated agent tokens. Unlike a standard card token, an agent token carries a usage policy:
who may use it, for how much, in which categories and over what period.
How an agent token works
User → sets the policy in the banking app
↓
The bank issues an agent token
↓
The agent receives the token (never the card details)
↓
The agent makes a purchase → the store
↓
The bank checks the purchase against the policy → authorization
The agent never sees the real card credentials. If it tries to step outside the policy — an amount
above the limit, a blocked category — the bank declines the transaction automatically.
Where this sits in the agentic protocol stack
| Layer | Instrument |
|---|---|
| Discovery and negotiation | MCP, A2A |
| Order protocol | ACP, AP2, YCP, UCP |
| Payment security | Shared Payment Token, Payment Mandate |
| Payment infrastructure | Visa Intelligent Commerce, Mastercard Agent Pay |
The card networks operate at the bottom layer — infrastructure. Protocols such as ACP sit a level
above, in order logic. A retailer does not need to understand tokenization internals: from where they
stand it all looks like ordinary acquiring.
Tip for retailers: the first step towards agentic commerce is not a payment integration — that
part works on its own — but a structured catalogue and support for an agentic protocol (ACP/YCP).
The payment layer will follow through your acquirer.