AI + payments

The agentic checkout stack

Market structure · 8 min read

When software becomes the shopper, the visual checkout page matters less. The underlying jobs do not disappear; they get separated into machine-readable services.

Discovery

The agent needs accurate catalog, availability, price, shipping, and merchant policy data. Structured merchant feeds become more valuable than visual merchandising for this use case.

Identity and authority

The system must know who the agent represents and what it is allowed to buy. A payment credential alone may prove that funds can move, but not necessarily that a purchase was authorized under the user's rules.

Payment and routing

The agent selects a rail based on cost, acceptance, rewards, credit, finality, geography, and merchant preference. Cards, account-to-account rails, and stablecoins can coexist because they optimize for different constraints.

Settlement

This is where stablecoins can materially compress time and cost, especially for machine-native and cross-border transfers. But settlement is only one layer of the checkout stack.

Post-purchase

Returns, refunds, disputes, delivery failures, and merchant support are economically important. Any agentic commerce architecture needs a way to represent these state changes after the initial payment succeeds.

The likely winners are not simply the companies with the cheapest rail. They are the platforms that control high-value orchestration points: identity, merchant acceptance, routing, risk, or the system of record.