Visa Wants to Be the Tollbooth for Agent-to-Agent Payments

Visa wants the next payments fight to happen before a human ever reaches checkout. On its earnings call this week, the card network told investors that AI agents, software that can shop and pay on a user's behalf, will start breaking purchases into many tiny transactions and even pay for their own data or computing costs one event at a time. If that happens, the company collecting trust signals and payment credentials for those machine-to-machine purchases gets a new tollbooth.

That is a more specific claim than "AI will change commerce." Visa chief executive Ryan McInerney said on the call that agents could create "an entirely new category of commerce with microtransactions," while also describing Visa as the "leading hyperscaler of payments globally," according to AlphaStreet's transcript of the company's fiscal Q2 2026 earnings call. Our read is that Visa is trying to move the conversation away from chatbot shopping demos and toward the infrastructure layer that decides which automated buyer is legitimate, where its payment token lives, and how many times it can safely spend.

Visa is not starting from a slide deck. Its Trusted Agent Protocol documentation describes signatures that are specific to a merchant and purpose, time-limited, and designed so they cannot be replayed or forwarded elsewhere. That matters because most merchants still treat bot traffic as abuse until proven otherwise. If autonomous software is going to buy flights, software subscriptions, or API calls, someone has to provide a way for a merchant to tell the difference between a fraudulent script and a user-authorized agent.

The company has also published a rough proof of concept for the machine-buyer world it is pitching. On the Visa CLI site, Visa describes a command-line payment tool for AI agents that can pay for image generation APIs, music generation endpoints, and proprietary data feeds without preloading API keys. That is not mass-market evidence. It is, however, a real artifact showing what Visa thinks agent commerce will look like: not one big checkout button, but software making small authenticated purchases as it works.

Visa says it already has some early traffic. In a December 2025 press release, the company said it was working with more than 100 ecosystem partners, that more than 30 partners were building inside the Visa Intelligent Commerce sandbox, that more than 20 agents or enabling companies were integrating directly, and that those collaborations had produced hundreds of controlled real-world agent-initiated transactions. The useful phrase there is controlled. That is more real than a mockup, but it is still a long way from proving that merchants broadly want machine customers showing up in production.

The competitive context is already here. Digital Commerce 360 reported earlier this month that Stripe expanded support for network-led agentic payment capabilities including Visa Intelligent Commerce and Mastercard Agent Pay, and that Google's Universal Commerce Protocol included both networks too. So this is not a story about Visa discovering AI. It is a story about payment networks, processors, and platform companies trying to define the trust layer before agent shopping becomes normal enough to standardize around somebody else's rules.

That is the pressure point. If agents really do start splitting purchases across merchants, timing windows, and per-call software services, then the winner is not necessarily the company with the slickest consumer interface. It may be the one that convinces merchants to accept software buyers without treating every automated action like fraud. Visa has enough technical scaffolding to make that ambition credible. It does not yet have enough independent public usage evidence to prove the new category is underway. The next thing to watch is whether these protocols escape Visa's own demos and controlled pilots, or whether agent commerce remains a standards war with very little traffic.