The battle to control payments may be moving to a place consumers never see. As AI agents begin searching, comparing and eventually purchasing on behalf of users, the decision over how to pay could happen before a customer reaches a checkout page, opens a digital wallet or even thinks about which card to use. That creates an uncomfortable possibility for banks, card issuers and wallet providers: years spent competing to become the customer’s preferred payment method may matter less when an algorithm is making the choice. In agentic commerce, the most valuable position in payments may no longer be top-of-wallet. It may be top-of-algorithm.
Payment Choice Is Moving Upstream
The traditional payment journey contains a clear point of decision. A customer reaches checkout, sees the available options and chooses between a card, wallet, bank transfer or another payment method. Even when the process is highly digital, the customer remains the final decision-maker.
Agentic commerce changes that sequence because the payment method can become part of the agent’s broader optimisation problem.
Imagine asking an AI assistant to book a hotel in Singapore for less than AED 1,500 per night, close to a particular district and with flexible cancellation. The agent could compare hundreds of properties, calculate the effective cost after loyalty benefits, select the room and determine that one card provides better travel rewards while another offers superior insurance. The customer may never consciously reach a payment-selection screen.
This is no longer only theoretical. Mastercard recently argued that card choice is beginning to shift earlier in agent-assisted commerce, with issuers potentially needing to make rewards, fees, protections and other benefits understandable to authorised AI agents. Visa is similarly building infrastructure that allows agents to participate in commerce, while Mastercard and Trip.com this week demonstrated an AI-powered travel booking experience designed to move from discovery through to consumer-authorised purchase.
The implication is significant. Payments may increasingly be selected before checkout even begins.
Top-of-Wallet Could Become Top-of-Algorithm
Banks have spent enormous amounts of money trying to influence payment behaviour. Rewards programmes, cashback, airport lounge access, merchant offers, instalment plans and premium card benefits all have a common objective: give the customer a reason to reach for one payment credential instead of another.
But AI agents do not reach for cards.
They evaluate information.
That creates a different competitive environment. A payment product may have excellent benefits, but those benefits become less valuable in an agentic transaction if the AI cannot discover, interpret and compare them. Mastercard has described this emerging requirement as making card value machine-readable, permissioned and contextual so an authorised agent can evaluate it against the customer’s preferences.
This could turn something banks traditionally treated as marketing into infrastructure.
A card offering 3% cashback on travel, zero foreign transaction fees and complimentary insurance may need to expose those benefits in a structured format that an agent can understand instantly. Eligibility conditions, merchant restrictions, reward caps and expiry rules may eventually need to become as machine-readable as payment credentials themselves.
The winning payment product might therefore not simply be the one customers remember. It could be the one their AI understands best.
Loyalty Does Not Disappear, but It Changes
That does not mean brands, rewards or loyalty suddenly become irrelevant. Customers will still have preferences, and those preferences can become instructions given to an agent.
Someone might tell an assistant to always use a particular airline, prioritise a favourite hotel programme or use a specific bank card whenever the difference in price is insignificant. Another customer might instruct the agent to maximise cashback regardless of provider.
What changes is how those preferences are executed.
Instead of repeatedly persuading a human customer at the point of purchase, payment providers may increasingly need to ensure that their value proposition survives translation into an algorithmic decision.
This could make loyalty simultaneously more powerful and less forgiving. A genuinely valuable benefit could be applied consistently across every eligible transaction because the AI never forgets it. A weak benefit could be ignored just as consistently because the AI is less susceptible to habit, prominent placement or the friction of switching payment methods.
The Wallet Could Become Infrastructure
Digital wallets have become one of the most strategically important interfaces in modern payments because they sit close to the transaction. They store credentials, authenticate customers and increasingly provide access to offers, identity and other financial services.
Agentic commerce does not necessarily eliminate wallets. It could make them less visible.
An AI agent may still use a tokenised card stored in a wallet, but the consumer may rarely open that wallet to select it. The wallet becomes part of the execution infrastructure underneath the agent rather than the interface through which the consumer makes the decision.
A similar shift has happened elsewhere in technology. Consumers routinely use infrastructure they rarely see, from cloud computing to payment gateways. Agentic commerce could push parts of consumer payments in the same direction.
This matters because control of the interface has historically provided enormous strategic value. Whoever controlled the checkout, wallet or merchant relationship had an opportunity to influence the transaction. If AI becomes the new interface, some of that influence moves with it.
Payment Providers May Have to Compete for Machines
The deeper change is that payment providers could eventually find themselves serving two customers simultaneously: the human who owns the money and the machine helping decide how it should be spent.
Finnoex recently explored this emerging problem in When AI Agents Start Spending Your Money, Who Is Actually the Customer?, where the central issue was delegated authority. Once an AI can legitimately transact for a customer, banks and networks need mechanisms to determine what the agent is authorised to do and whether its actions remain within the customer’s original intent.
The industry is already moving in this direction. Mastercard, Visa and Ant International announced a joint initiative in September to develop common standards for identifying and verifying AI agents participating in commerce. Mastercard’s Agent Pay framework similarly includes registered agents, tokenised credentials and mechanisms intended to verify consumer intent.
Trust therefore becomes intertwined with payment selection. An agent needs to know not only which payment option produces the best economic outcome, but which credentials it is authorised to use, under what circumstances and with what protections.
That makes the emerging contest bigger than another battle between cards, wallets and bank transfers.
It is a competition to become selectable by machines.
Checkout May Become the Least Interesting Part of the Transaction
This also challenges one of the industry’s long-standing assumptions: that checkout is where payment competition happens.
In an agentic environment, much of the important work could happen before the merchant ever requests payment. Product selection, merchant selection, financing, rewards optimisation, currency choice and payment routing could all be considered within the same automated decision.
Accenture has estimated that more than 30% of online commerce could run through AI agents by 2030, while arguing that payment choice could increasingly become algorithmic as agents optimise across fees, rewards and customer preferences. That scale is far from guaranteed, but the direction of current infrastructure investment suggests payment companies are preparing for a meaningful change in how transactions originate.
Banks therefore face a strategic question that is subtly different from the one they have spent years answering.
It may no longer be enough to ask: how do we become the customer’s preferred payment method?
The next question is: how do we make sure the customer’s AI reaches the same conclusion?
What it means for the industry
- Top-of-wallet could become top-of-algorithm. Payment providers may increasingly compete to be selected by AI agents rather than directly by consumers at checkout.
- Card benefits will need to become machine-readable. Rewards, fees, protections and eligibility rules may need to be structured so authorised agents can evaluate them automatically.
- Loyalty could become more rational. AI agents may apply genuinely valuable benefits more consistently while ignoring offers that provide little measurable advantage.
- Wallets may become less visible without becoming less important. They could remain critical credential and authentication infrastructure while the AI agent becomes the primary customer interface.
- Trust and discoverability will converge. Payment methods will need to be both understandable to agents and demonstrably authorised for agent-initiated transactions.
- Banks may need to market to algorithms as well as people. The next generation of payment competition could depend on whether a financial product’s value can be discovered, interpreted and selected by machines.
