Visa Powers AI Shopping Agents With ‘Intelligent Commerce’ Payment Rails

Visa wants the next wave of agentic artificial intelligence (AI) to do more than curate a dream vacation or the perfect outfit to wear for it. It wants it to pick up the tab.

Unveiled Wednesday (April 30) at the company’s Global Product Drop, the Visa Intelligent Commerce program opens the network’s rails to developers building AI agents that search, recommend and now pay on behalf of consumers.

The stakes are considerable. Generative AI (GenAI) platforms are influencing what people buy. The missing link is a payment mechanism that is both invisible to consumers and accepted, quite literally, by merchants.

“This is going to transform shopping and buying — we’re letting AI developers and engineers use the Visa network to allow AI agents to find, and buy, on [the consumer’s]  behalf in a seamless and safe way,” Mark Nelsen, Visa’s global head of consumer products, told PYMNTS’ Karen Webster on the eve of the announcement.

At the center of the effort is an AI-ready card, a credential developers can spin up through a bundle of Visa application programming interfaces (APIs). As Nelsen put it: “The APIs will have an AI-ready card. In a hypothetical search for the cheapest flight to Cancún, for example, a traveler uploads an existing Visa credential, while in the background Visa authenticates the cardholder with payment passkeys, tokenizes the 16-digit number and seats the token inside the AI agent.”

That behind-the-scenes swap shields account data, eliminating the clunky step of keying in card numbers every time the agent hops to a new site.

It also takes card-on-file commerce to an entirely new, contextually powered ecosystem, allowing consumers to leave their physical and digital wallets at home.

Nelsen walked through an inventory of five modules that developers can cherry-pick or adopt wholesale. Modules, he said, are built on three decades of Visa’s work in using AI to fight fraud. Visa says its AI engines blocked roughly $40 billion in fraud last year. Visa Intelligent Commerce imports the same models into agent-driven transactions.

• Authentication: Confirms the AI agent is authorized to act for a specific consumer, extending identity verification to AI commerce.
• Tokenization: Replaces card data with network tokens that work anywhere Visa is accepted while keeping sensitive details under wraps.
• Payment Instructions: Lets the user preset dollar limits, merchant categories or real-time approval prompts — guardrails that load directly into VisaNet.
• Personalization: With the shopper’s consent, the module shares basic spend patterns so the agent can rank offers by preferred airline, hotel-price comfort zone or dining habits.
• Signals: Streams transaction signals back to Visa in real time to trigger risk controls and aid dispute resolution.

Nelsen emphasized that, together, the tools aim to make AI-powered buying as routine and secure as tapping a phone at checkout.

But here’s where it can get interesting and where AI agents and AI cards can simplify the complexity of shopping.

GenAI chatbots can already present a short list of hiking boots or boutique hotels. What they can’t yet do is close the sale without punting the consumer to a checkout page. Or in the case of a complex search that asks for the most stylish hiking boots for the summer trip to Cinque Terre, and what restaurants are best to watch the sunset, there are multiple searches, steps and a big gap from information and intent to purchase. That gap, Nelsen argued, is where cart abandonment and lost revenue lurk.

“We ask the consumer … whether they would like to share information around their spend behavior with the agent so that it knows their preferences, such as the airline you prefer and the typical amount of money that you spend on hotels. ‘Where do you like to go for dinner?’” he said. “That’s personalization.”

Once a shopper sets, for instance, a $500 ceiling for a hotel or an airline ticket, the AI agent scours sites, picks seats or rooms and pays, constrained by the limits embedded in the token created by the AI card.

“[Visa] verifies your instruction,” Nelsen said, “and loads that instruction into VisaNet, and the agent will work on their behalf. … If everything matches, we’ll send the transaction to the issuing bank to be approved.”

Nelsen said that a consumer has all the security, control and protections in place that come with using their Visa card, even though it is the AI agent that is “making” the transaction.

The same tokenization that reassures consumers also signals to merchants that the purchase request is legitimate, not a bot fraudster. Users can require the AI agent to ping them for approval above a certain spend, or to pause before purchasing from unfamiliar sellers.

Unlike fledgling AI startups, Visa doesn’t have to lay new rails. Its tokenization framework already covers more than 200 countries and territories, and the Intelligent Commerce APIs are live now for partners ranging from Anthropic and OpenAI to Microsoft, Samsung and Stripe. “Any Visa credential can work,” Nelsen said, because the underlying technology has been field-tested for years.

That global footprint means an AI agent could book a table at an impossible-to-get-into Manhattan restaurant — or a last-minute villa on Italy’s Lake Como — without exposing card data or tripping fraud filters.

Merchants, bruised by fraud losses and rising customer-acquisition costs, want conversions. Consumer want friction-free checkout. And regulators want proof that AI won’t introduce new risk. Visa’s bet is that sliding its fraud models and dispute-resolution tools under the hood will let all three constituencies adopt AI commerce faster.

“It’s not only safe for the consumers,” Nelsen said. “It’s safe for the whole ecosystem, for the banks and the merchants as well. We’re excited about what the program will enable.” And, he added, “search used to be simple, and then it got complicated — now we can make it personalized and simple at the same time.”

Visa isn’t predicting how quickly consumers will surrender purchase decisions to software agents, but it is positioning itself as the payments backbone when they do. By turning card numbers into tokens, layering on spend controls and piping in real-time fraud defenses, Intelligent Commerce tries to ensure the AI agent economy is born with same adult-grade security that exists in the digital and mobile ecosystems today.

Developers can start coding today. Pilots are expected to surface in the coming months. If the model sticks, shopping bots could soon move from novelty to necessity, with Visa swiping the AI-powered card.

The post Visa Powers AI Shopping Agents With ‘Intelligent Commerce’ Payment Rails appeared first on PYMNTS.com.