Agents will spend from stored cards. Here are the controls, consent and wallet-sync checks card issuers need before agentic payments scale.
Agentic payments put software, not a person, at the point of purchase. The card behind it still belongs to an issuer.
The controls that matter are specific: which agent, which merchants, how much, and when to ask the cardholder.
Wallets are becoming the place those credentials and controls live, which makes provisioning and lifecycle sync an issuer problem.
What are agentic payments? Agentic payments are payments an AI agent makes on behalf of a person or business, within limits they have set. The agent starts the purchase and the cardholder stays accountable. For a card issuer, that means the card, its controls and its audit trail have to work when the one using it is software.
In November 2026 Google launched its wallet app on iPhone in the US, 13 years after its first attempt, which it later abandoned. The app stores payment cards and loyalty credentials. It doesn't offer tap-to-pay on Apple devices, a limit Google framed with the phrase "at this time."
What Google emphasized instead was agents. P.J. Linarducci, Google's vice president of consumer payments, said the wallet will be "a secure hub for managing agentic purchases across Google." That refers to Google's own agents.
The wider pattern is easier to see across platforms. OpenAI launched a ChatGPT wallet for agentic payments on November 13, 2026, and Meta's Muse agent uses Stripe Link for checkout. Analysts told Payments Dive much the same thing in different words. Laura Huntley of FTI Consulting called Google's move "an agentic commerce play." GlobalData's Yasaswini Pujitha said that as agents take on more online shopping, the focus is "likely to shift from the physical terminal to stored credentials and user controls." IDC's Aaron Press said "agentic enablement is a significant factor." Google's head of payments, Stavan Parikh, told Payments Dive in an earlier interview that the wallet is being scaled "for end-to-end commercial journeys."
Digital wallets are multiplying across operating systems and platforms. For an issuer, each one is a place where a card has to be provisioned, tokenized, monitored and kept accurate through its life. When a copy drifts out of date, cardholders see a declined payment and call support. When an agent is involved, the same drift shows up as a failed task that the cardholder blames on the card.
A card limit is a start. Agent spending usually needs conditions: which agent, which kind of merchant, what amount, over what period, and when to pause and ask the cardholder.
In our conversations with issuers and payments teams, three themes keep coming up: consent, auditability and human oversight. Consent means the cardholder has agreed, in terms they understand, to what the agent can do. Auditability means anyone can reconstruct what happened and why. Human oversight means a person can see activity as it happens and step in. If people can't see what an agent did and stop it, they're unlikely to delegate spending to one.
Card networks and platforms are building their own consent and control tooling, and for some programs that will be enough. The gap tends to show where an issuer needs rules that are its own: its risk appetite, its limits and its audit trail.
Wallet coverage. Know which wallets and devices your cardholders can add a card to, and where support stops.
Token-level visibility. See activity per token, not only per card, so you can tell which wallet or agent a payment came through.
Real-time agent limits. Enforce amount, merchant and frequency rules at authorization, not after the fact.
Lifecycle sync. Make sure a replaced, frozen or expired card updates in every wallet right away.
A readable audit trail. Give cardholders and your disputes team a clear record of what an agent bought and under which permission.
If the honest answer to any of these is "not sure," that's a sensible place to start.
A finance team gives an agent a card to buy API usage and software credits. The rules: approved suppliers only, up to $2,000 a month, and a human approval for any single purchase above $500. The agent can act quickly inside those lines and has to stop at the edges. Someone has to enforce that at authorization, and someone has to be able to explain the record afterwards.
Known: Google has launched its wallet on iOS in the US without tap-to-pay. OpenAI and Meta have launched wallet or checkout routes for agents. Platforms describe wallets as hubs for agent purchases.
Not yet known: how quickly agent spending scales, which protocols win, how responsibility for a dispute is split between the agent provider, the issuer and the merchant, and when wallets like Google's reach other countries.
It's tempting to picture consumer shopping first. We expect the early cases to be B2B and pay-per-use, where rules are clearer, amounts are predictable and the value is easy to measure. Consumer retail is likely to follow once trust, consent and recourse are well understood. This is a view about direction, not a statement of what any program can do today.
Our direction at Thredd is to work at the infrastructure layer for agentic commerce: provisioning, tokens and controls that let a wallet, an agent and an issuer trust one another. If you're an issuer thinking through your own answers to the checklist above, talk to our team.
Sources
Payments Dive, "Google wallet returns to Apple," November 2026: https://www.paymentsdive.com/news/google-wallet-returns-to-apple/875182/
Talk to our team about provisioning, tokens and controls for agentic payments.