AI is already changing the way people shop. Consumers are increasingly asking artificial intelligence to research products, compare prices and narrow down choices that would previously have required multiple searches, websites and reviews. The next step is much more significant: allowing those same systems to make the purchase.
Mastercard recently predicted that more than 300 million online shoppers globally could routinely use AI agents to shop and pay by 2030, equivalent to more than one in ten online shoppers. That forecast shows how quickly agentic commerce is moving from experimentation and towards genuine consumer behaviour. But the more important question is not around if the technology can make a payment but what needs to happen before consumers are comfortable handing over that level of authority.
Shopping with AI is not the same as paying with AI
There is an important distinction between asking an AI agent to find the best pair of running shoes for a particular budget and allowing it to spend the money required to buy them. Visa CEO Ryan McInerney recently observed that adoption of AI for shopping is happening, while autonomous payments have yet to follow at the same pace. Consumers may be increasingly comfortable using AI to influence a purchasing decision, but moving from advice to financial authority represents a very different trust threshold.
That difference matters because recommendations can be rejected and search results can be ignored. Once an agent makes a payment, however, money has moved and contractual relationships between consumer, merchant and payment providers have been created. At that point, agentic commerce stops being purely an AI challenge and moves into a payments challenge.
We can authenticate an agent, but can we prove intent?
Much of the technology needed to enable AI agents to transact is developing rapidly. Payment credentials can be protected, agents can be authenticated and transactions can be subjected to increasingly sophisticated controls. That is important progress, but authentication and intent are not necessarily the same thing.
Imagine that a consumer instructs an agent to buy a particular product for no more than £100. The agent finds an alternative for £120 that it judges to be better value and completes the purchase. The agent itself may be legitimate, the payment credentials may be valid and there may be no fraud in the conventional sense. Yet the outcome has exceeded what the consumer believed they had authorised.
The same problem becomes more complex when an agent buys from the wrong merchant, misunderstands a preference, purchases the wrong quantity or acts on information that is no longer accurate. Traditional payment systems have largely been built around interactions where the individual making the purchase and the individual authorising the payment are one and the same. Agentic commerce introduces another participant into that relationship, acting with delegated authority, which makes proving the boundaries of that authority crucial.
Consent needs to become more specific
We are already seeing technology companies think more carefully about those boundaries. Meta’s recently announced personal AI agent, Muse, for example, has been designed to seek approval before sensitive actions such as making a purchase, while its payments integration can use a single-use virtual card scoped to an approved transaction. The initial service is launching in the US, but the principles it demonstrates have much wider relevance for how agentic commerce could develop.
These safeguards point towards a broader issue. Simply granting an agent permission to “shop for me” is unlikely to be sufficient as autonomous commerce becomes more sophisticated. Consumers may want to define how much an agent can spend, which merchants it can use, what categories it can purchase, whether substitutions are acceptable and when it must return to the individual for approval.
The payments ecosystem will therefore need ways of recording not simply that consent existed, but what was actually consented to. That distinction could become fundamental when something goes wrong, because proving that an agent had permission to act is not necessarily the same as proving that the eventual transaction matched the consumer’s instructions.
Accountability cannot become an afterthought
Agentic transactions potentially involve several parties, including the consumer, the AI provider, the merchant, the payment provider and the card scheme or payment network. If the transaction does not reflect the consumer’s instructions, responsibility may not always sit neatly with one party.
There will not necessarily be one simple answer and different circumstances will produce different responsibilities. Existing consumer protection and dispute mechanisms will therefore need to evolve alongside the technology, with clearer standards around delegated authority, consent and redress. Those questions need to be addressed as agentic commerce develops, not after autonomous transactions have already reached significant scale.
That is why an AI agent cannot be treated as simply another checkout channel. It changes who is making decisions within the transaction, how those decisions are authorised and what evidence may be needed if the outcome is later challenged.
Trust will determine how quickly agentic commerce scales
The technical ability for an AI agent to make a payment is only one part of the equation. For agentic commerce, or agentic shopping, to become mainstream, consumers will need to understand when an agent can act independently, when their approval is required and what protections exist if the outcome does not match their instructions. Merchants will similarly need confidence that an agent arriving at checkout is legitimate, has the authority to transact and represents a genuine customer. Therefore, success will lie making autonomy understandable, with clear boundaries around consent, control and accountability.
The real test of agentic payments will be whether the industry can demonstrate why a payment was made, what the consumer authorised and where accountability sits when the result does not match their intent. Getting that framework right will be fundamental to earning the trust agentic commerce needs to scale.
