The term "AI payments" is used in several different ways. This page separates them into categories so the rest of the AI Payments Map has a clear and stable starting point.
What all five categories share is the presence of software — an AI agent, an automation, or a business process — in the initiation or approval of a payment. They differ by payment infrastructure, by who is paying, by risk profile, and by regulatory environment.
1. Consumer agentic checkout. A person asks an assistant to buy something, and the assistant completes the purchase on a consumer-facing surface. Associated with ChatGPT Instant Checkout, Gemini in Search, and the agentic programs from Visa, Mastercard, American Express, and PayPal. Rails: cards and wallets. Paying party: the cardholder.
2. Enterprise accounts payable and spend automation. A company uses software — increasingly including AI copilots — to process invoices, approve payments, and send money to suppliers. Rails: ACH, SWIFT, SEPA, corporate cards. Paying party: the company or the finance team. The category itself is not new; what is changing is the AI component and the approval logic.
3. Machine-to-machine and API payments. One software system pays another — for an API call, compute, inference, or a dataset. Protocols such as x402 (see the protocols overview) were designed for this. The rails are most often stablecoins on Layer 2 networks, although the model does not require crypto in principle.
4. Stablecoin and crypto flows initiated by software. Treasury operations, supplier payouts, and cross-border settlement initiated by agents or automations and settled in stablecoins. Distinct from M2M: the amounts are larger, the counterparties are real-world entities, and compliance obligations shape the design.
5. Programmable treasury and conditional payments. Payments that execute only when predefined conditions are met — delivery, project milestone, or an oracle signal. Historically this has been the domain of smart contracts; it is becoming more relevant for central bank digital currencies and for AI-managed treasury operations.
In each category, software is present at or before initiation and makes decisions that used to be made by a human at a keyboard. The infrastructure underneath was designed around the opposite assumption, and the resulting gap appears differently in each category.
A buyer evaluating an "AI payments" product is usually looking for a tool for one of these five categories, not all of them at once. A product that claims to solve all five deserves closer examination. The rest of this knowledge base is organized around that distinction.