Visa and Mastercard Just Joined a 26-Company Club Trying to Figure Out How AI Agents Should Pay for Things

Somebody had to organize the AI agents before they started spending money on their own. On August 18, stablecoin infrastructure company Rain took that job, launching the Agentic Payments Alliance (APA) with more than two dozen companies attached to it on day one, including, notably, both Visa and Mastercard, who don't share a room together very often.

The full roster reads like someone invited the entire payments industry to one meeting: Visa, Mastercard, Fiserv, Circle, Solana, Remitly, Avalanche, Chainalysis, Fireblocks, Uniswap Labs, Turnkey, Shift4, and a dozen more, 26 founding members in total. Solana Foundation confirmed its membership directly, posting on X that "agents are becoming economic actors, and how they pay for things needs to be on a global money layer that stays open to anyone."

Why now?

Rain's pitch is basically: nobody has answered these questions yet, and someone was going to answer them badly if nobody got together first. "No single company should get to decide how agents transact on someone's behalf," said Rain co-founder and CEO Farooq Malik. "That has to come from the platforms building the rails, the regulators setting the rules, and the innovators closest to how agents are actually being used today."

Mastercard's Sherri Haymond, who runs digital commercialization there, put the urgency a different way: "The risk in a moment like this is not that the industry moves too slowly, it's that innovation outpaces alignment." That's a polite way of saying twenty companies are all about to build incompatible versions of the same thing unless they agree on some ground rules first.

The rules in question aren't small. How does a merchant know an AI agent is actually authorized to spend a customer's money? What does fraud detection even mean when the "customer" doesn't type, doesn't have a location, and doesn't behave anything like a human shopper? Who owns the loyalty points when an agent, not a person, makes the purchase? None of that has an answer yet, which is exactly the gap the APA says it wants to close.

What Rain Actually Brings To This?

Rain didn't show up to this empty-handed. The company spent the past year building an Agent Control Layer and something it calls Scoped Cards, payment credentials that let an AI agent spend, but only within limits a human sets first. That work is apparently what gave Rain enough credibility to convene the rest of the industry instead of just building its own walled-garden version of agent payments.

Structurally, the APA isn't Rain's property. It's described as a working coalition governed collectively by its founding members, who'll set the charter and mission themselves rather than take orders from whoever launched it. Early priorities: shared research, testing agent identity and authorization standards, and lobbying regulators before those regulators write rules without industry input. Founding members also get first access to Rain's Agentic Startup Program, an accelerator whose first cohort of five startups will demo directly to Alliance members.

Zil Bareisis, a director at the payments research firm Celent, compared the effort to EMVCo, the body that got Visa, Mastercard, Amex, and the rest of the card networks to agree on chip-card security standards decades ago. It's not a perfect analogy since EMVCo came from an existing industry with existing rails, while agentic commerce is still figuring out what "the transaction" even looks like. But the instinct is the same: better to hammer out shared standards early than clean up a mess of incompatible ones later.

This Is The Third Or Fourth Land Grab This Year, Not The First!

Here's the part that makes the APA interesting rather than just another press release: it's not alone. Google has its own Agent Payments Protocol (AP2), an open rulebook for verifying agent-initiated transactions, and Mastercard is a backer there too, sitting on two competing standards bodies at once. Coinbase, meanwhile, launched the x402 Foundation under the Linux Foundation with Cloudflare and Stripe as co-founders, pulling in Google, Microsoft, AWS, and, again, both Visa and Mastercard. And Stripe's own Tempo blockchain went live earlier this year alongside the Machine Payments Protocol, co-authored with Tempo and backed by launch partners that include Anthropic, OpenAI, DoorDash, Shopify, and once more, Visa and Mastercard.

I keep coming back to that pattern. The same handful of payment giants are hedging their bets by joining every agentic-payments coalition that forms, which either means they genuinely don't know which standard wins yet, or means none of these "alliances" is really about consensus. It's about not getting locked out of whichever one does. Probably both.

Circle's presence in the APA fits its obvious interest: as the issuer of USDC, more stablecoin-denominated agent commerce is directly good for Circle's business. Solana's involvement signals which blockchain ecosystems want a piece of that transaction volume once it starts flowing at scale. Neither company is hiding the motive, and there's nothing wrong with that, but it's worth naming rather than pretending everyone joined purely out of civic duty.

The Number Everyone Keeps Citing

McKinsey's projection — $3 trillion to $5 trillion in global agentic commerce by 2030 — shows up in literally every announcement about this space, the APA's included. It's a big number, and also the kind of forecast that's impossible to verify five years out and easy to repeat anyway. Worth treating as a directional bet rather than a fact: something in payments is about to change because of AI agents, and nobody, including McKinsey, actually knows the size of it yet.

What's real right now is smaller and more concrete: an agent identity problem, a fraud-detection problem that doesn't map to existing tools, and a regulatory vacuum that 26 companies would rather fill together than have filled for them. Whether the APA ends up mattering the way EMVCo did, or ends up as one of several competing standards that never quite converge, is genuinely an open question. For now, it's one more sign that the payments industry has decided AI agents spending money isn't a hypothetical anymore.