The Age of AI That Buys for You: Visa and Mastercard Moved on the Same Day

Search, compare, add to cart, and pay — the whole sequence is now being handled not by a human but by AI, and it has moved past the pilot stage. On June 10, 2026, Visa announced a partnership with OpenAI at its own forum in San Francisco, letting agents complete payments on a user's behalf inside ChatGPT. The same day, Mastercard unveiled a framework for transactions between machines. The two largest payment networks moved on the same day, as if by appointment. Agentic commerce is no longer a question of "when," but of "how to prepare."

Two payment giants moved on the same day

On June 10, 2026, at the Visa Payments Forum, Visa announced its OpenAI partnership built around "Visa Intelligent Commerce." Visa's payment capabilities are embedded into OpenAI products including ChatGPT and the coding agent Codex, so that even agent-initiated transactions carry tokenized credentials, real-time authorization, and fraud monitoring. Visa makes its own network — which processes over 300 billion transactions a year — directly available. It also launched, the same day, "Agent Scoring" to rate agents, an "Agentic Registry" for identity, and a model for large transactions.

By coincidence, on that same June 10, Mastercard announced "Agent Pay for Machines." Its focus is less on buying on a person's behalf and more on automating the micro- and continuous transactions that machines exchange in the background. The two firms' centers of gravity differ slightly, but they point the same way: handing AI a "wallet" safely.

Why "the agent buys"

Grasp the scale and the urgency makes sense. Morgan Stanley Research estimates that agent-led shopping could capture $190–385 billion of U.S. e-commerce spending by 2030 — a 10–20% share in the base and bull cases (Morgan Stanley, as of December 2025). Its analyst Nathan Feather sums up the shift in a line:

The evidence is there. Per the same research, about 23% of Americans have already purchased something via AI in the past month. Roughly half of large language model users used them to research or compare prices, and 30–40% actually made a purchase (Morgan Stanley, as of December 2025). Groceries and CPG lead and hold the largest growth runway over the next five years. The entrance to shopping is quietly shifting from the search engine and the app to the conversation.

Tokenization solves the "handing AI your wallet" problem

Here is where everyone tenses up: safety. Give an AI your card number and who knows what it gets used for. Visa's answer to that fear is tokenized credentials. In place of a raw card number, it issues a network token bound to a specific agent and a specific use. A token issued for grocery shopping cannot book travel; a token capped at $200 cannot authorize a $500 charge. These are not contractual promises but technical constraints enforced at the network layer, in real time, at the moment of authorization.

Users can also pre-set conditions — spending limits, merchant categories, whether approval is required — and the agent can only act within that frame. Visa supports multiple standards — the Trusted Agent Protocol, Machine Payments Protocol, Agentic Commerce Protocol, and Universal Commerce Protocol — refusing to privilege any single agent ecosystem. Rather than fence off a standard, it carries any of them onto its network: the world of payments is retracing the lesson MCP taught — interoperability wins.

From the shelf that "gets searched" to the shelf that "gets chosen"

For sellers, this changes the premises of acquisition and merchandising. Products have long competed to be "searched" by human eyes. But when the agent is the buyer, a product first becomes something to be read, compared, and chosen by AI. More than the cleverness of a banner or headline, what decides whether you make the shortlist is whether your structured product data — inventory, price, shipping terms — can be retrieved accurately by a machine.

Visa's infrastructure is the payment layer; the contest is decided just upstream of it. How well can you present, in machine-readable form, the materials an agent uses to decide "buy here" — accurate product information, clear return terms, trustworthy reviews? This is continuous with GEO (generative engine optimization): preparing information to be cited by AI and preparing products to be chosen by AI use the same muscle.

Preparing for an age judged by AI

So what should sellers anchor to? Before rushing into steps, fix the lens to look through.

First, hold your product data in two layers: a human-facing description and a machine-facing set of facts. Whether you can offer price, stock, specs, shipping, and returns as unambiguous structured data is the precondition for entering an agent's consideration set.

Second, speak about your payment-safety design as a promise to customers. Controls like spending limits and approval requirements are backend technology, but they are also the grounds for "nothing gets bought without my say-so." The customers most anxious about delegated purchasing are exactly the ones who deserve a clear explanation of how it works.

Third, prepare to observe agent-led sales as a distinct metric. Who bought, from which conversation, under what conditions — this leaves a different footprint than traditional web analytics. Sellers who build that measurement early will read the direction of change sooner.

Payments are infrastructure; the contest is upstream

That Visa and Mastercard moved on the same day means the foundation of agentic commerce is largely set. But the more the foundation is settled, the more the difference is made on top of it. If payment becomes safe and seamless, consumer choice shifts to "whose agent experience is trustworthy and precise." What is tested is not appeal aimed at people, but honesty aimed at machines. Is your product ready to be read correctly — and chosen — by a machine?

Key takeaways

On June 10, 2026, Visa partnered with OpenAI on "Visa Intelligent Commerce" and Mastercard unveiled "Agent Pay for Machines" the same day, setting the payment foundation for agent-led purchasing. Morgan Stanley projects agent-led shopping could reach 10–20% of U.S. e-commerce spending ($190–385B) by 2030 (as of December 2025), and about 23% of Americans have already bought via AI. Safety rests on tokenized credentials (use and limit fixed in technology) plus user pre-settings, and Visa supports multiple industry standards. The seller's contest is upstream of payment — whether you can present product data accurately and machine-readably, ready to be "chosen" by AI.