Agentic Payments

Your AI Agents Now Pay Each Other

Mastercard's Agent Pay for Machines lets autonomous AI agents transact with each other directly. When no human approves each payment, your only control is a spending mandate you can verify.

By Leon Soliman · 2026-06-30 · 2 min read

What Mastercard actually switched on

For most of the past year an AI agent could prepare a transaction, but a person still pressed pay. Mastercard's Agent Pay for Machines removes that step for machine-to-machine commerce. One agent can now pay another agent directly, in amounts as small as a fraction of a cent, for things like incremental access to data or services. Adyen, Coinbase, and Cloudflare are among the first partners.

The design choice that matters is where the rules live. Instead of holding each agent's spending permissions in a private database, Mastercard records them on public blockchains, so any party in a transaction can check that an agent is acting within the authority its owner granted. The point is not the technology. The point is that authorization has moved from a human in the loop to a rule that travels with the agent.

Why this is a different risk than an agent with a card

An agent that buys on your behalf still has one principal and one beneficiary: you. An agent that pays other agents is part of an economy that runs between machines, where transactions happen faster than anyone can review them and the counterparties are themselves automated. The failure mode changes from a single bad purchase to a chain of small, automatic payments that drift outside what you intended.

The mistakes of an agent that can only answer cost you an awkward sentence. The mistakes of an agent that can pay move money before you know a decision was made. That is the line a company crosses the moment it connects an agent to a payment rail.

What to settle before you connect an agent to money

Decide the mandate first: what an agent may spend, on what, up to what limit, with which counterparties, and for how long. Make the mandate machine-readable and verifiable, not a paragraph buried in a policy document. Keep a second control for anything above a threshold, and a kill switch that revokes authority instantly.

Treat every agent as a non-human identity with its own credentials, its own budget, and its own audit trail. The firms that govern this well will not be the ones that move slowest. They will be the ones that can prove, at any moment, exactly what each of their agents is allowed to do.

Frequently asked questions

Does Agent Pay for Machines mean my company has to use cryptocurrency?

No. The protocol uses public blockchains to record the permissions a human grants an agent so they can be verified by any party, not to force payment in crypto. The relevant question for a business is governance: can you define and prove what each of your agents is authorized to spend.

We are not deploying paying agents yet. Why act now?

Because the controls have to exist before the capability is switched on. Once an agent can move money, a missing spending mandate or a slow revocation process is no longer a policy gap, it is an open account. Setting the boundary early is far cheaper than reconstructing it after a disputed transfer.

The era when a human approved every payment is ending. What replaces it is not trust in the agent, it is a mandate you can set, verify, and revoke. The companies that write that boundary now will own their automation. The ones that wait will read its limits in their own ledgers.

AI Agents Payments Governance NonHumanIdentity