Quick Answer: Visa has said it is working with more than 50 partners to prepare for AI agent payments, alongside plans for stablecoins and cyber risk set out at Singapore Fintech Festival 2026. A Visa survey of consumers in Singapore, Australia and Japan found generative AI use ranging from 67% to 90%, but only 20% to 30% of respondents would trust AI to buy everyday household goods. The reporting describes a direction and a partner effort rather than a live consumer product.
1. What Visa actually put on the table
The reported facts are narrow. Visa set out plans at Singapore Fintech Festival covering three areas: AI payments, stablecoins and cyber risk. Separately, the company has gathered more than 50 partners to prepare for AI agent payments.
That is an announcement of direction plus a partner programme. The source material does not say which partners have signed, which markets are involved, what volumes are being tested, or what a shopper would see at checkout. Anyone judging how close agent-initiated payments are to daily use needs to hold those gaps in view.
Read plainly, the effort is about assembling the plumbing and the rulebook before demand shows up. Cards have absorbed new purchase channels before, from e-commerce to in-app and contactless, and each time the hard work sat in the rules about who carries the risk when something goes wrong.
2. How an AI agent would actually pay
An agent in this context is software that completes a task on a person's behalf without step-by-step instructions. Reordering a usual grocery basket, for example, or buying a replacement part once a specification is fixed.
The payment problem is that the human is not present at the moment of authorisation. A card used online today assumes someone typed, tapped or approved at that point. An agent needs authority granted in advance and expressed as something the payment system can enforce: spending ceilings, permitted merchant categories, a time window, a frequency cap, and a credential the agent can use but cannot freely copy elsewhere.
Tokenisation already points this way. A token replaces the underlying card number with something scoped to a device or a merchant and can be suspended. Agent payments push the same idea further, because the credential has to carry scope rather than simply stay secret. The open question that follows is liability: who answers for a duplicate order, a wrong item, or an agent that misreads a size chart.
3. High use of AI, low willingness to hand over the wallet
Visa's survey covered consumers in Singapore, Australia and Japan. Reported use of generative AI ran from 67% in Japan to 90% in Singapore. When asked whether they would trust AI to buy their everyday household goods, only 20% to 30% said yes.
The two numbers are worth reading together. Familiarity with the tool is not the same as consent to spend. Everyday household goods are close to the least risky purchase a person can delegate, and even in that case roughly seven in ten respondents in these three markets declined.
Two limits apply. The survey covers three countries, not a global sample, and the source does not reproduce the question wording, which shapes answers on trust. Treat the figures as a signal about hesitation in high-income Asian and Australian markets rather than a universal reading.
4. Why stablecoins appear in the same conversation
Visa grouped stablecoins with AI payments and cyber risk when setting out its plans, and the pairing is telling. Stablecoins are tokens designed to hold a steady value, usually one US dollar, and they move on public blockchains rather than through correspondent banking rails.
For an agent that transacts on its own, that matters in two practical ways. Transfers can settle at any hour, including weekends and holidays when traditional settlement sits idle. They can also carry conditions in code, so a payment can be made to depend on a delivery confirmation or a price threshold.
The source does not detail what Visa intends to do with stablecoins, so the connection should be read as context rather than a roadmap. Stablecoin settlement also brings its own obligations: reserve and redemption quality, a route back into local currency, and screening for sanctions and illicit flows. Those controls do not disappear because software initiated the transfer.
5. Fifty partners is a coordination problem
A partner count is a measure of how many parties have to agree before an agent payment works end to end. Issuers, acquirers, merchant platforms, wallets and identity providers each hold a piece of the same transaction, and each has its own rules.
The questions that decide whether this becomes usable are mostly unglamorous:
- How a credential expresses scope, and how a customer or a bank revokes it quickly.
- Who absorbs the loss when an agent buys the wrong item or buys it twice, and whether existing dispute rules can stretch to cover it.
- How screening and fraud checks run when the initiator is software acting for a named person.
- Whether a merchant can tell an agent transaction apart from an ordinary online payment, and price or route it differently if needed.
6. What would show this is working
Progress will be visible in specifics rather than partner totals. Named pilots with published limits, rules that say who is liable when an agent errs, and merchant tooling that routes agent traffic distinctly would all be stronger signals than a longer list of collaborators.
The cyber risk thread belongs here too. Every agent holding spending authority is a target worth attacking, and the aggregation is new: a single compromised agent configuration could place many small payments that individually look unremarkable. Visa naming cyber risk alongside the other two topics suggests it sees the same exposure.
For now the honest summary is that the industry is preparing rules and connections ahead of proven demand. Whether shoppers want this is unresolved, and the survey Visa itself commissioned is the clearest evidence that many do not yet.
For businesses on the merchant side, agent payments would shift attention from accepting a payment to controlling one. A wallet that can act on its own needs limits, revocable permissions and a clear audit trail, and those requirements land on whoever holds the funds and settles them. Paycot works on crypto acquiring, an on/off-ramp across major networks and fiat currencies, and OTC services for larger orders, which is the layer where settlement choice and transaction monitoring meet. Teams weighing how automated or agent-driven traffic would fit their own setup can map the network and fiat options that match their volume before committing to a design.
Planning how your business would handle automated or agent-initiated crypto payments? Talk to the Paycot team about networks, fiat settlement and OTC for orders above €10K.
This article is editorial coverage based on the reporting listed in Sources. It is not financial, legal or tax advice, and the plans described may change or be withdrawn.
Sources
- https://fintechnews.sg/141553/singapore-fintech-festival-2026/visa-singapore-fintech-festival-2026/
