AI shopping agents are beginning to move beyond product recommendations and controlled demonstrations.
Visa has announced that AI agents can now complete purchases on behalf of consumers at participating merchants’ websites across Europe. The transactions take place in live commercial environments rather than test storefronts, bringing agentic commerce closer to everyday use.
During these transactions, AI agents can browse products, choose suitable options, and initiate a purchase according to instructions set by the cardholder. Visa connects the consumer’s bank, the merchant, and the AI system while providing authentication and payment infrastructure.
The development addresses one of the hardest questions in agentic commerce: how can an online store know that an AI agent is legitimate and authorized to act on behalf of a real customer?
Most AI shopping tools currently help consumers research products, compare alternatives, or narrow down their choices. The user still completes the final purchase.
Visa’s live transactions introduce a different model. Consumers can define what they need, set parameters, and let an AI agent handle more of the shopping process.
The participating merchants include lastminute.com, Frasers, Cleverbridge, and BrickDepot. The trials span travel, retail, and e-commerce, showing that the model is not limited to a single product category or transaction type.
More than 30 European issuing banks are involved in the initiative, including Barclays, BBVA, Commerzbank, ING, Klarna, Revolut, and several other major financial institutions.
Allowing AI agents to access merchant websites creates new security and operational challenges.
Automated traffic is not always welcome. Retailers already need to distinguish customers from bots used for scraping, fraud, or other unwanted activities. Agentic commerce adds another category: automated systems that genuinely represent a shopper and intend to make a purchase.
Visa’s Trusted Agent Protocol and Agent Directory aim to provide merchants with a consistent signal that identifies a verified AI agent. Retailers can then decide how that agent may access their website, interact with product listings, and complete transactions.
Importantly, Visa says merchants do not need to rebuild their entire infrastructure. The verification signals can work alongside existing risk controls, policies, and user experience systems. Cloudflare and Akamai are supporting implementation at the infrastructure level.
Trust also depends on proving that the customer authorized the transaction.
Visa Payment Passkeys connect each agent-initiated payment to a verified user and an explicit instruction. The system is designed to support European Strong Customer Authentication requirements while keeping the cardholder in control of when and how a purchase happens.
This distinction matters. An AI agent should not have unrestricted freedom to spend. It needs clear limits, such as a maximum budget, preferred product characteristics, approved merchants, or a defined purchasing task.
Agentic commerce will struggle to gain acceptance unless consumers, merchants, and banks can see who authorized each action and why.
Payments may complete the transaction, but product information helps the agent decide what to buy.
An AI agent comparing products needs accurate specifications, prices, availability, images, compatibility details, delivery information, and merchant policies. Vague or inconsistent listings make it harder for the system to determine whether a product meets the customer’s instructions.
For brands and retailers, this raises the value of structured product content. Product information must serve human shoppers while also becoming understandable to AI systems that browse, compare, and select products independently.
In agentic commerce, incomplete data may affect more than the appearance of a product page. It may determine whether an AI agent considers the product at all.
Visa’s announcement does not mean that autonomous shopping will become standard immediately. Consumer habits, merchant adoption, regulation, and technical standards still need time to develop.
However, live transactions represent a meaningful step beyond AI shopping demonstrations. The payment, authentication, merchant recognition, and consumer-control layers are beginning to work together in real commercial settings.
For e-commerce businesses, the preparation goes beyond adding an AI chatbot. Retailers will need secure systems, clear policies, machine-readable product information, and reliable transaction data.
The next online customer may still be a person. Yet the system that browses the catalog and places the order could increasingly be an AI agent acting on that person’s behalf.
Read further: News, agentic commerce, AI, e-commerce, ecommerce, Icecat, product content, Visa
