Mastercard's latest pitch for agentic commerce starts with an attractive premise: software agents will create new payment use cases and could transact far more frequently than people. The company enters that market from strength. American Banker reported second-quarter net revenue of $9.3 billion, up 12%, and 18% currency-neutral growth in value-added services. Yet management also called agentic commerce an early market.
The distinction matters. More automated transactions do not automatically create a profitable new rail. Before an agent can pay, every participant needs to know whose authority it carries, what it is allowed to buy, which party bears an error and how value settles. Mastercard's opportunity is to make that authorization chain portable across merchants and payment types. Its risk is that open protocols make the same chain available without making Mastercard indispensable.
An agent needs a mandate before it needs a card
A human at checkout supplies context that payment infrastructure often takes for granted. The buyer chooses the item, confirms the amount and authenticates the transaction. An autonomous agent separates those steps in time and may act repeatedly without a person watching. A credential alone cannot show whether a purchase remained within the user's instruction.
Mastercard's original Agent Pay framework addressed that gap by registering and authenticating agents, using tokenized payment credentials, identifying agent-facilitated transactions and preserving rules over what an agent may purchase. It also described authentication and a process for clarifying unfamiliar transactions. The product is therefore as much an identity-and-consent layer as a payment button.
Agent Pay for Machines extends the idea to software buying digital resources from other software. Mastercard says the system can credential each agent, attach verifiable intent, enforce spending limits and settle across cards, accounts and stablecoins. The economic moat would come from making those assertions recognizable to issuers, merchants and processors at scale. If a legitimate agent looks like an ordinary bot, fraud controls may decline it; if an unauthorized agent looks legitimate, the network inherits a loss and trust problem.
Transaction density is not automatically revenue density
Mastercard's official product announcement envisions continuous, high-frequency payments, including amounts worth only fractions of a cent. That expands the addressable number of transactions, but it also changes their economics.
A payment that is tiny cannot carry the same absolute processing, fraud, support and dispute cost as a retail purchase. Machine commerce therefore rewards infrastructure that can make authorization and settlement almost entirely programmatic. Batching, netting or stablecoin settlement may help in some use cases, while guaranteed settlement and existing card protections may matter more in others. Mastercard supports multiple rails, which lets it route by use case rather than insist that every machine payment behave like a card purchase.
Still, no public disclosure shows Agent Pay for Machines pricing, transaction volume, loss rates or revenue. The more than 30 named participants and supporters demonstrate ecosystem interest, not proven unit economics. Nor did Mastercard separately quantify agentic-commerce revenue in the earnings figures reported by American Banker. For now, agentic payments are an option on future transaction density, not a measurable earnings segment.
Open protocols make distribution a shared asset
The competitive field reinforces that uncertainty. Stripe's Machine Payments Protocol lets agents answer an internet payment request with stablecoins or fiat methods, including cards and buy-now-pay-later credentials. Stripe says merchants can receive the money through the same balance, reporting, tax and refund infrastructure they already use. This lowers the integration cost for its installed merchant base.
Visa's Trusted Agent Protocol uses agent-specific cryptographic signatures to communicate the agent's intent, consumer recognition and optional payment information. Visa explicitly designed it to align with other protocols and existing web standards. Mastercard also describes its approach as open and interoperable.
Openness expands distribution because agents, merchants and processors do not want a different identity language for every network. It also limits proprietary control. If a common intent credential can be routed through several networks and rails, competition shifts toward acceptance, risk models, settlement reliability, dispute handling and price. Mastercard can benefit because it already connects issuers and merchants globally, but the protocol layer alone may not create lock-in.
The strongest counterargument to Mastercard's position is therefore not that cards disappear. It is that cards become one selectable settlement method behind an agent interface controlled by someone else. A merchant, processor or agent platform could choose the cheapest suitable rail for each transaction. In that world, Mastercard must earn routing share through authorization quality and services rather than rely on consumer card choice at checkout.
The scoreboard is authorization quality
The useful milestones are operational. Growth in credentialed agents matters only if merchants recognize them and issuers approve them at attractive rates. Rising transaction count matters only if fraud, disputes and servicing costs stay below the revenue those transactions generate. Multi-rail support matters only if it improves completion, settlement speed or economics without confusing responsibility when something goes wrong.
Evidence that would strengthen the moat thesis includes disclosed merchant adoption beyond pilots, higher authorization rates for verified agents than for unidentified automated traffic, low dispute and fraud rates, repeat machine-payment volume and revenue that is incremental to ordinary card processing. Evidence against it would be widespread adoption of competing open credentials with little difference in approval or risk performance, or machine payments migrating mainly to cheaper rails where Mastercard supplies no differentiated service.
Mastercard's advantage today is not ownership of the agent. It is its position between the user mandate and final settlement. If it can make that chain auditable, portable and economically efficient, agentic commerce can deepen the network's role even when the payment is not a card. If those functions become standardized commodities, the same openness that helps the market grow will make value capture much harder.