Three bitter rivals just agreed on the rules for AI agents that buy things

On September 10, three payment companies that normally spend their days taking business away from each other sat down to write shared rules. Ant International — the global arm of Alipay's parent, Ant Group — joined Visa and Mastercard, a trio with more than $77 billion in combined annual revenue, to build what they call a "Know-Your-Agent" interoperability framework, aiming for a common way for card networks, digital wallets, and marketplaces to verify an AI agent acting on a person's behalf.

The premise underneath it is the next leg of the AI cycle I keep pointing to. AI is moving from something that recommends products to something that buys them. The companies behind the framework project that autonomous agents will orchestrate $3 trillion to $5 trillion of global consumer commerce by 2030. That is the point where years of compute spending finally convert into a revenue layer — and whoever owns the rails that money flows across collects a toll on every machine-initiated purchase.

That's the real subject here. Not the protocol acronyms, but the commercial logic of three competitors agreeing to cooperate — and the tension hiding inside it.

The generation gap in the payment rails

Scrappy me-too agreements aside, each company already had its own answer to the same problem: how do you certify that a machine buying on my behalf is actually authorized to do so. Visa had built its Trusted Agent Protocol, Mastercard its Verifiable Intent, and Ant its open-sourced Agentic Mobile Protocol (AMP). A patchwork of standards is the enemy of adoption, so the KYA framework is the attempt to make those systems recognize each other's trust signals, pin every agent to a validated operator or cardholder, and keep monitoring it after the fact.

Here is where the architecture gap shows up, and it's the part to hold onto. Ant's protocol did not bolt an AI checkout onto the old system. It was built for the workload of machine-driven commerce, with a settlement mechanism designed for agent-to-agent transactions as small as a fraction of a cent — down to $0.000001. Traditional card rails, engineered around human purchases and an interchange fee of roughly two to three percent, are structurally mismatched to ultra-high-frequency micro-transactions. There isn't enough fee on a sub-cent purchase to make those rails pay.

That's the tension beneath the cooperation. Visa and Mastercard are not being charitable. By co-building the trust layer with Ant — keeping every agent tethered to a validated cardholder — they pull the agent economy back into their own fee rails instead of letting it route around them. Cooperation here is moat defense, not fellowship. Both companies' valuations sit on the assumption that the toll booth survives the machine age.

What is real, and what is a talking point

Discipline matters on the evidence here, because the two parts of this story are not the same thing. AMP is genuinely delivered — open-sourced in April, connected to more than 4.4 billion digital wallet users through Alipay+, with pilot relationships at Visa and Mastercard. The KYA framework announced this week is a framework: the companies have agreed on principles, not a technical standard, with no release timetable, no pilot volumes, and no revenue disclosed. The stocks barely moved on the day — the market read it for what it is, early positioning rather than an earnings event.

There is also a structural problem for the retail investor that the headlines skip over: there is no Ant International ticker. Ant Group is private, its long-stalled IPO still a rumor. The two ways to own these rails are Visa, at roughly 29 times trailing earnings, and Mastercard at about 30 times — both extraordinary businesses, with revenue growing in the mid-teens, operating margins in the high-fifties to low-sixties percent, and price-to-sales ratios around 14 to 15.

So the honest framing is not "buy because of this announcement." It is a watching point on the central question of the card networks' moat: when machines do the spending, does the money flow through the cards, or around them? The single evidence boundary that separates those outcomes is whether agent-orchestrated payment volume starts appearing on the card networks' own reported volumes — versus wallet-to-wallet agent flows that never touch a card account. That is the number that tells you whether a moat is extending into the agent economy, or quietly being bypassed.