When your buyer is an agent

Somewhere in your 2026 plan there is a line about agentic commerce. An agent that books travel for your customers, restocks a client’s warehouse, negotiates a software renewal. The payments industry has stopped debating whether these agents will transact and started arguing about how. 

‍

Visa has opened a single integration for agent-initiated payments and completed secure AI transactions with partners. Mastercard has brought Agent Pay to market. A third camp argues agents should skip card rails entirely and settle in stablecoins. None of them disputes the part that lands on you. Someone still has to issue the credential, control the spend and settle the money.

‍

Most of the noise is about the buying experience. Underneath every agent transaction sits the same machinery that sits under a card tap today: a regulated issuer, a scheme membership, authorisation logic that decides in milliseconds whether this payment, at this merchant, for this amount, should happen, and settlement that moves real money between real institutions under real supervision. If you are building a product that agents will buy through, that machinery is now your problem.

‍

Agents raise the stakes on that machinery rather than replacing it. A human cardholder notices when something is wrong. An agent executes. So the controls cannot live in the buyer any more; they have to live in the infrastructure. The credential itself becomes the control. It works only for that agent, at that merchant, up to that amount, inside that window. This is the same issuing discipline that already lets an insurer settle a claim onto a card that works at one retailer for one purchase, and a fleet operator authorise fuel at one terminal at one moment. Agentic commerce applies that logic at machine speed.

‍

The rails question, cards or stablecoins, is probably a false choice. Agents will use both: tokenised card credentials where acceptance and consumer protection matter, and stablecoin settlement where speed and programmability matter. What matters is whether both ends sit inside one regulated framework, because a compliance failure executed by an agent is still a compliance failure, at a much higher cadence, in your name.

‍

Our view is unfashionably calm. Agentic commerce is coming, the schemes are building the standards, and the hard problems are the ones payments has always had: identity, authorisation, control and settlement. So put one question to your infrastructure partner: When my buyer is an agent, can your controls keep up? That is the layer we build, and we would rather be early on the controls than early on the hype.

‍

If agents are anywhere in your plans, start with the controls.

‍

‍

‍

‍