The Maze: Mastercard thinks AI agents will change who clicks “buy,” but not necessarily who controls the money. During its second-quarter call, CEO Michael Miebach argued that cards can prevail because the network already connects banks, merchants, credentials, data and risk controls. Software shoppers will not carry plastic, but every autonomous purchase still needs a trusted record of who authorized the agent, what it may buy and who is liable when something goes wrong.
Mastercard is turning the checkout into an authorization stack. Agent Pay gives an AI agent a tokenized credential rather than raw card details. Verifiable Intent binds the transaction to a user’s permission and creates an auditable record. Bank authentication, fraud screening and authorization still decide whether money moves. Mastercard then adds data signals and merchant tools around the transaction. It wants to sell both the rail and the traffic lights.
The machine version goes beyond consumer shopping. Agent Pay for Machines handles continuous, low-value transactions between software systems. A business can credential an agent, set limits, let it buy data or computing services and settle across cards, bank accounts or stablecoins. Autonomous commerce may look less like one €80 basket and more like hundreds of tiny background purchases. More than 30 participants or supporters joined the launch—but that is an ecosystem list, not 30 disclosed paying customers.
The model has crossed from demo to production, within controlled boundaries. Worldline, ING and Mastercard completed an agent-initiated payment involving an ING cardholder and a Dutch merchant. The agent searched for concert tickets within a budget, but the customer explicitly approved the purchase. ING authenticated and authorized it; Worldline processed both issuer and merchant sides; Mastercard supplied network rules and identifiers showing that an agent was involved. It proves the plumbing can work. It does not prove consumers are ready to delegate shopping at scale.
The revenue opportunity sits above basic processing. Mastercard’s quarterly net revenue rose 12% to $9.3 billion, while value-added services grew 18% on a currency-neutral basis. Identity, tokenization, fraud controls, intent records, data signals and dispute evidence all fit that faster-growing services layer. The same products also defend the network against pay-by-bank systems, AI-platform wallets and stablecoins that could route around cards. Mastercard is hedging rather than choosing one rail: its machine-payment design supports cards, accounts and stablecoins, while management still expects cards to remain central.
Why it matters: Agentic commerce shifts power from the visible checkout toward infrastructure that makes a machine’s decision trustworthy. Merchants need structured product data, machine-readable policies and payment systems that can distinguish a legitimate agent from automated fraud. Banks and networks preserve control if they own identity, permission and liability. AI platforms may gain the interface but still rent the financial trust layer. The next scoreboard is real transaction volume, fraud and chargeback performance, merchant conversion, service pricing and which rail agents choose when no human is watching.
