Within a single week, Visa and Mastercard both staked a claim to the same future, and it is not a coin or a wallet. It is the plumbing for money that moves itself. Both companies are positioning to be the layer that banks and, increasingly, AI agents plug into when value changes hands, which tells you where they think the next decade of payments is decided.
Two announcements, one land grab
At its Visa Payments Forum, Visa unveiled a set of AI, stablecoin, and digital-token initiatives it frames as the foundation for intelligent, programmable commerce, as we covered this week. The package includes a technology layer for tokenised deposits, letting banks turn ordinary deposits into programmable, always-on digital money rather than a stablecoin or a central-bank currency, expanded stablecoin support that Visa says is reshaping the back end of commerce, and a tie-up with OpenAI for agent-driven payments. The detail on Visa’s own newsroom makes the ambition plain: Visa wants to be infrastructure, not an issuer.
Mastercard moved in parallel. Its Agent Pay for Machines, reported across the trade press including Finextra and CCN, targets always-on, machine-to-machine payments for AI-led transactions, and coverage from blockchain outlets noted it leans on onchain verification and stablecoin settlement. Strip the branding and the two announcements rhyme: both networks are building rails for transactions that no human clicks to approve.
From cards to coins to agents
The move fits a long pattern. The networks first digitized the card, then tokenized it so a real card number never touches a merchant, and they have spent years turning physical-world trust into API calls. Stablecoins threatened to skip that machinery entirely, letting value move bank-to-bank, or wallet-to-wallet, without the networks in the middle. Tokenised deposits are the counter: keep the money inside the regulated banking system, but make it programmable enough to compete with a stablecoin on speed and composability. Agents are simply the next abstraction on top, a buyer that never sleeps and never types a card number, and whoever the agent calls to settle becomes the new default.
Why now: the decision layer is moving
The urgency is easier to understand through a point made by Paymentology chief technology officer Tim Joslyn to PYMNTS: the rails already work. Machine-to-machine payments, automated billing, and API-consumption models have existed for years. What is changing is that AI is now the thing making the decision. As commerce moves from human-speed browsing to machine-speed orchestration, the competitive battleground stops being the checkout button and becomes the trust infrastructure underneath it: identity, authorization, and AI-native fraud control. In Joslyn’s framing, power shifts away from whoever owns the storefront and toward whoever owns the decision layer.
The competition is everyone at once
What makes this contest unusual is that the networks are fighting on several fronts simultaneously. Stablecoin issuers want to be the settlement asset. Large banks want to defend their deposit franchise and the customer relationship that comes with it. Model providers such as OpenAI want to own the agent that initiates the purchase, which is why Visa’s tie-up with it matters more than a logo on a slide. Visa and Mastercard are trying to be indispensable to all of them at once, the neutral layer every party routes through. It is the same strategic position they hold in card payments, extended to a world where the buyer is software.
The skeptic’s case
Programmable money has been promised before, and adoption is not guaranteed. Tokenised deposits require banks to coordinate on standards they have historically been slow to agree on, and the economics only work at scale. Regulators are watching closely: the Financial Stability Board has opened consultation on the responsible adoption of AI in financial services, a signal that the supervisory framework for agent-initiated payments is unsettled. And the hardest unsolved problem is liability. When an autonomous agent authorizes a payment that turns out to be fraud or error, who is responsible, the user, the bank, the network, or the model, remains an open question that no press release this week answered.
What it means for the finance leader
For banks and treasurers, the strategic question is no longer which token wins. It is who supplies the connective layer when settlement goes on-ledger and machine-initiated. Tokenised deposits are the banks’ attempt to answer stablecoins without surrendering the customer relationship, and the networks are offering to be the rails that make that answer work end to end. That is a genuinely attractive bargain, but it carries an old risk in new packaging. If the networks own the agent-authorization and settlement layer, they collect a toll on machine-initiated commerce the same way they do on card swipes today.
What to evaluate now
Audit whether your payments roadmap still assumes a human at checkout, because that assumption is the one being removed. Ask your providers how agent identity and authorization are established, and where liability sits when an autonomous agent transacts. Assess stablecoin and tokenised-deposit settlement readiness rather than dismissing both as crypto noise. And scrutinize tokenised-deposit arrangements for counterparty and settlement risk, asking the blunt question the marketing will not: does this reduce risk, or simply route the same exposures through a new intermediary wearing a familiar logo.
The week’s signal is not that any single product will win. It is that the incumbents have decided agent-initiated money is close enough to fight over now, and they are building the rails before the standards, the regulators, or the challengers settle the question. Finance leaders who wait for that dust to clear may find the toll booths already built.
