Agentic Payments Explained: How Mastercard and Visa Are Letting AI Shop

An Agent Gets Its Own Card, Not Yours

The core design decision behind both Mastercard's Agent Pay and Visa's Intelligent Commerce is the same: an AI agent never touches a person's actual card number. Instead, the network issues a fresh, scoped virtual card tied to a specific task — a spending limit, an approved merchant category, sometimes an expiration tied to the task itself — and hands that virtual card to the agent to complete the transaction. If the virtual card is ever compromised or misused, the exposure is limited to whatever scope it was issued with, not a person's whole account.

That mirrors the confirmation-and-scoping pattern showing up across the broader personal-agent category this year — Meta's Muse pausing before a purchase, Gemini Spark asking before spending — but at the payment-network level rather than inside a single company's product.

From Task to Completed Purchase

The Real Barrier Isn't the Technology

Both Mastercard and Visa built the scoping and virtual-card mechanics specifically because the biggest obstacle to agentic commerce isn't technical — it's trust. Surveys through 2026 have consistently found that a majority of consumers remain uncomfortable letting an AI agent complete a purchase entirely on its own, even with spending limits in place. The infrastructure existing doesn't guarantee adoption; it just removes one of the excuses for not building it.

Dispute and Refund Mechanics Still Apply

Because agentic payments run through the same underlying card networks as any other transaction, existing consumer protections — chargeback rights, fraud dispute processes, purchase protection benefits tied to a card — generally still apply to a purchase an agent completes, since the transaction itself looks like an ordinary card payment from the network's and issuer's perspective. What's genuinely new is the added layer of scoping before a transaction happens: a spending limit and merchant restriction set in advance function as a form of prevention on top of the dispute mechanisms that already existed, rather than replacing them.

You Don't Need an Agent to Benefit From This Infrastructure

Even for someone with no interest in letting an AI agent shop on their behalf right now, the existence of this infrastructure matters indirectly: it's what will eventually make agentic shopping features in mainstream apps and assistants feel like an ordinary card transaction rather than a novel, higher-risk way to pay. As more retailers and payment processors adopt these scoped virtual card standards, the practical experience for a shopper who does decide to try an AI agent for a purchase should increasingly resemble using a normal card with a spending limit, rather than something that requires trusting an entirely new payment mechanism.

The Core Facts, in One Place

To recap: Mastercard's Agent Pay, built with Alchemy, and Visa's Intelligent Commerce both let an AI agent complete a real purchase using a scoped virtual card rather than a person's actual card details, with spending limits and merchant restrictions set in advance. The technology already exists and works through standard card-network rails, meaning existing consumer protections generally still apply — the main open question is consumer comfort and adoption, not whether the payment infrastructure itself functions.

Common Questions