An AI payment agent is software, usually built on a large language model, that can complete a purchase or transfer funds on a person’s or business’s behalf without a human clicking “buy” for each transaction. This is distinct from an AI shopping assistant, which searches, compares, and recommends products but hands the final checkout step back to a person. An agent completes that step itself, within limits the user has set in advance.
How an Agent Gets Permission to Pay
Agentic commerce depends on three layers: identity, mandate, and settlement.
Identity establishes that a given piece of software is a recognized agent and not a bot scraping a site or attempting fraud. Visa’s Trusted Agent Protocol, launched in October 2025 with Cloudflare, addresses this by helping merchants distinguish legitimate agents from malicious traffic.
Mandate is the permission structure: a user-defined scope covering what the agent can buy, from which merchants, up to what amount, and for how long. Google’s Agent Payments Protocol (AP2) formalizes this as a cryptographically signed mandate that travels with the transaction, so a merchant can verify the agent is acting within the limits its owner set. Mastercard’s Agentic Tokens work similarly, binding a tokenized card credential to a specific agent, a specific merchant scope, and a specific consent policy, rather than exposing the raw card number.
Settlement is how the money actually moves once identity and mandate are verified. This is where the industry has split into two camps.
Card Rails vs. Stablecoin Rails
Card networks are extending existing infrastructure. Mastercard’s Agent Pay, announced in April 2025 with Microsoft, IBM, and Braintree as launch partners, issues Agentic Tokens from its Digital Enablement Service. Visa has paired its Trusted Agent Protocol with a stablecoin settlement pilot that reached a $7 billion annualized run rate by April 2026, according to Visa’s own disclosures, and the company has said it is working with Coinbase to align on interoperability.
Coinbase took a different route with x402 protocol launched in May 2025 that revives the dormant HTTP “402 Payment Required” status code, reserved in the original web specification but never implemented, to settle payments in stablecoins directly over HTTP.
Card rails typically carry interchange of roughly 1.5% to 3.5% plus a per-transaction fee that can run $0.50 to $0.80, figures the Federal Reserve Bank of Kansas City has estimated for credit transactions. That economics does not work for an agent paying a fraction of a cent per API call or data query — the kind of machine-to-machine transaction agentic commerce is expected to generate at volume. Stablecoin rails on chains such as Base, Arbitrum, and Solana settle in seconds at sub-cent cost, which is why x402 has gained traction specifically for agent-to-agent and agent-to-server payments rather than consumer retail.
The split narrowed on July 14, 2026, when the Linux Foundation launched the x402 Foundation, an open-governance body with 40 founding members including Visa, Mastercard, American Express, Stripe, Google, AWS, and Coinbase, to steward the protocol jointly rather than leave it as a single company’s product.
What This Means for Merchants, Banks, and Consumers
For merchants, agentic commerce means adapting checkout flows to authenticate software rather than only people, and deciding which settlement rails to accept. For banks and card networks, it means a choice between defending existing interchange economics on higher-value consumer purchases and ceding the sub-dollar, high-frequency machine-to-machine segment to stablecoin rails. For consumers, it means the mandate layer becomes the practical safeguard: the value of an AI payment agent depends on how precisely its spending limits and permissions can be defined and verified, not on how convincingly it can shop.
Market forecasts vary widely by methodology. Juniper Research projected $8 billion in global agentic spend for 2026, scaling to $1.5 trillion by 2030, while McKinsey’s QuantumBlack practice has put 2030 orchestrated retail spend at $3 trillion to $5 trillion. Both are early-stage estimates for a category still being defined by the standards bodies and protocols described above, and both will likely be revised as adoption data accumulates.
