Real Card Rails Carry Visa and BBVA Agentic Payments

On Thursday (July 2), BBVA said an AI agent-initiated transaction had used real card credentials and an active merchant’s systems, a small test with a large message: agentic payments may be able to start on today’s card rails rather than wait for a new payment network.

The Spain-based financial group completed the transaction with Visa, with the agent acting on behalf of a cardholder, according to PYMNTS. The timing matters because Visa launched its Agentic Ready program in Europe in March to let banks test AI-led payments in production-grade environments with selected merchants.

This is where the hype gets a useful constraint. The test doesn’t prove that fully autonomous shopping agents are ready for mass use. It does show that the hard parts of payments, credentials, authentication, fraud checks, issuer oversight and merchant acceptance, can be tested inside existing infrastructure.

July 2: BBVA’s first AI agent-initiated transaction moved through real payment infrastructure

BBVA said the payment was initiated by an AI agent on behalf of a cardholder. It used real card credentials and the systems of an active merchant, not a purely theoretical checkout simulation.

That distinction is the core of the story. Many AI-commerce demos can show an agent searching, comparing or recommending. The BBVA and Visa test moved into the payment step, where banks and networks care less about the chat interface and more about authorization, fraud, consent and accountability.

The transaction ran through Visa Intelligent Commerce, which Visa describes as its platform for agentic commerce. BBVA said the transaction used technologies already associated with secure digital payments, including tokenisation and real-time fraud monitoring. It also used Visa Payment Passkeys, a biometric authentication method for online payments, to support the Strong Customer Authentication requirements mandated by the European Union.

“Working with Visa through the Agentic Ready programme allows us to participate in the next phase of commerce, where AI agents can initiate transactions on behalf of cardholders using an infrastructure that already delivers scale, security and control,” Roberto Pagán, Head of Consumer Payments at BBVA Spain, said in the release.

The practical roles are clear. BBVA sits closest to the customer and the card account. Visa provides the network, token and commerce infrastructure. The merchant provides the real-world purchase endpoint. The AI agent initiates, but the payment still has to pass through familiar control points.

March to July: Visa Agentic Ready turns AI payments into a bank test case

Visa launched Agentic Ready in Europe in March, saying the program gives banks a structured way to test payments made by AI agents on behalf of consumers. In the first phase, participating banks run agent-initiated transactions in production-grade testing environments with selected merchants.

That framework matters because agentic payments are not just a user-interface problem. They force banks to answer a sharper question: when software acts for a customer, how does the system prove that the customer actually allowed the action?

Visa’s answer, based on the material disclosed, is to keep the transaction inside the card network’s existing control model while adding agent-specific context. Visa said at its Visa Payments Forum 2026 that Visa Intelligent Commerce provides “the trust, controls and connectivity needed for AI agents to securely discover, initiate and complete transactions.” It also described tools such as an Agentic Directory, Agent Score, token enhancements and a Large Transaction Model trained on billions of transactions to improve fraud detection and authorization performance.

For banks, this is less about letting a chatbot spend freely and more about defining the operating envelope. A user’s instruction, a credential, an authentication event and a merchant transaction all need to line up.

That control theme echoes other payments debates XOOMAR has tracked, including banks’ preference for certainty in real-time payments and the operational pressure created when liquidity decisions hit the send button in real-time payment systems. Agentic payments raise a different version of the same issue: the system needs confidence before money moves.

How agentic payments work when the shopper is no longer clicking checkout

In plain terms, agentic payments mean a person authorizes an AI agent to take a commercial action within limits. The agent might search, compare and initiate a purchase, but the payment still needs a credential, authentication, authorization and merchant acceptance.

A simplified flow looks like this:

• Intent: The cardholder tells the agent what outcome they want.
• Permission: The user defines what the agent is allowed to do.
• Selection: The agent identifies or confirms a purchase.
• Authentication: The customer or payment system validates the transaction as required.
• Authorization: The issuer and network apply fraud and payment controls.
• Acceptance: The merchant receives the payment through existing systems.

The reason card credentials matter is speed to adoption. Existing card rails already have tokenization, issuer authorization, merchant acceptance and dispute frameworks. A new payment method would need to recreate much of that before merchants and banks trusted it at scale.

But the design problem is unforgiving. The system must show that the agent is acting for the cardholder, within permitted boundaries, and not as a new path for fraud. If that proof breaks, agentic commerce becomes a risk channel instead of a convenience layer.

Spain’s consumer signal: 62% already use AI for shopping research, but payment is the harder step

BBVA’s release included a useful consumer data point. According to Visa data cited by BBVA, 62% of surveyed consumers in Spain use AI to look for gift ideas, research products and compare prices. The survey was prepared for Visa by Morning Consult and conducted between October 14th to October 28th, 2025, among 1,000 adults in each market.

That shows why payment networks are pushing now. Consumers are already using AI near the top of the buying funnel. The unresolved step is letting an agent cross from advice into payment.

A realistic future use case, not disclosed as the BBVA test itself, would be narrow and permissioned. A cardholder could ask an assistant to find a product or travel option, then allow a purchase only if it fits pre-set instructions. The payment system would still need to authenticate the cardholder, pass issuer checks and route the transaction to the merchant.

The missing piece is not imagination. It is permission design. Customers need to know when an AI agent can spend, how much control they retain, and how quickly they can shut it off.

The risks now shift from demo feasibility to consent, fraud and merchant readiness

The BBVA and Visa transaction shows technical feasibility. It does not settle the operating rules.

Three risk buckets now matter most:

Liability is the hardest commercial question. If an agent buys the wrong item, acts on bad product data or completes a purchase the user later contests, banks, merchants and networks will need rules that are clear before volume arrives.

Eduardo Prieto, Country Manager of Visa in Spain, framed Visa’s role around trust and existing protections.

“By connecting issuers, merchants and AI systems through our network, we are enabling this next phase of commerce using the infrastructure and protections already in place.”

The next decision point: narrow use cases before broad autonomy

Visa’s own comments suggest a gradual path. Michele Herron, senior vice president and head of North America Value-Added Services at Visa, told PYMNTS that personal assistants able to participate in commerce could take shape within the next three years, starting with simpler tasks such as reordering household items, booking routine travel or comparing subscription options.

That is the right order. Agentic payments will likely begin where intent is clear, merchants are trusted and the purchase rules are easy to express. More complex buying decisions require stronger controls and clearer accountability.

For now, the BBVA and Visa test lands a narrow but important point: agentic payments don’t have to begin outside the banking system. The rails may be ready before customer behavior, merchant operations and bank policies fully catch up. The next thing to watch is whether Visa Agentic Ready produces repeatable bank and merchant playbooks, not just isolated proof points.

Disclaimer: This XOOMAR analysis is for informational and educational purposes only. It is not financial, investment, legal, tax, or professional advice. It does not provide buy, sell, hold, price-target, portfolio, or personalized recommendations. Verify information independently and consult qualified professionals before making decisions.

The Bottom Line

• BBVA and Visa showed that AI agent payments can move beyond demos into real payment infrastructure.
• Using real card credentials and active merchant systems gives banks a practical path to test consent, fraud controls and accountability.
• The test supports the idea that agentic commerce may scale through existing card networks rather than requiring a new payments system.