Your AI assistant is about to have its own credit card. The companies deciding how that works didn't ask you.
In two weeks, Google, Mastercard, Visa, Stripe, and five others all launched "the standard" for AI agent payments. There are now six standards.
Something happened in the last two weeks that barely made the news, and it's going to affect how every piece of software spends your money for the next decade.
AI assistants got bank accounts.
Not metaphorically. Google partnered with the Solana Foundation to launch a service that lets AI agents pay for things on the internet, per request, without a login or a subscription. MoonPay and Mastercard launched a virtual debit card specifically for AI agents to spend money wherever Mastercard is accepted. A company called Oobit gave AI agents their own Visa corporate cards. And a payment protocol that Coinbase built, now handed off to the Linux Foundation as an "open standard," counts Google, Stripe, Mastercard, Visa, Microsoft, Amazon, American Express, and Shopify as founding members.
All of this in fourteen days.
The pitch is straightforward and genuinely useful: AI agents are increasingly doing real work on the internet, booking things, buying API access, running software, and they've had no clean way to pay for any of it. They've been borrowing your credentials, scraping around OAuth flows, pretending to be you. Giving them their own payment identity is the right idea.
Here's what's wrong with how it's being done.
When a new technology needs a standard, you get a standard. When every powerful company in an industry needs a standard, you get six standards, each named by a different company, each subtly incompatible with the others, all of them called "the open solution."
Count the "standards" that launched in the last few months for AI agent payments alone.
There's the protocol Coinbase built, now under Linux Foundation governance. Stripe launched its own competing version in March. OpenAI and Stripe collaborated on a third one focused on shopping. Visa has its own mandatory protocol for agents transacting on Visa's network. Mastercard is building another. Google quietly shipped a fifth one that uses a different authorization model from all the others.
That's five or six competing universal standards, depending on how you count Mastercard.
The flagship announcement of the week, the Google and Solana partnership, supports two of them. Not both at once. Either one. The biggest launch in this space in months shipped with a dropdown where the payment layer goes.
This isn't a technical problem. Engineers could make these protocols talk to each other by Tuesday. It's a political one. Every company with enough scale to matter has decided the correct move is to name their protocol the standard, get their logo on the press release, and quietly route as much transaction volume through their own infrastructure as possible.
The history of this move is long and the outcome is always the same.
When the internet itself was being built, the same war happened over email protocols, over web standards, over who controlled the plumbing. What resolved it wasn't cooperation. It was volume. Whoever got enough developers building on their protocol first made everyone else irrelevant by default. The winner captured the infrastructure. Then they charged for it.
That's what credit cards are. Visa and Mastercard didn't win because they had the best technology for moving money. They won because enough merchants accepted them that consumers carried them, and enough consumers carried them that merchants had to accept them. Today, every time you tap your card, a merchant pays roughly three percent of the transaction to that network. Forever. For doing nothing except being the standard that won in 1975.
The agent payments race is the same race, at AI speed, with the same companies in the room.
There's a question hiding underneath all this that the press releases don't touch.
When your AI agent books a flight, or renews a subscription, or pays for a cloud service on your behalf, who actually authorized that? Your agent has a card now. But did you set limits on it? Does it have a spending cap? Can it pay for things you'd never approve yourself if you saw them in advance? If it gets compromised, who's responsible?
The companies building the payment rails have a detailed answer for how the money moves. They have almost no answer for how the authorization works.
Full disclosure: one of the projects I'm currently building sits exactly in this gap. Not the payment network layer, the authorization layer. The thing that makes sure a human explicitly approved what the agent is about to do, before it does it. Working on it changes how you read these announcements. The infrastructure they're racing to build assumes the authorization problem is solved. It isn't.
That's a harder problem than building a card. It's also the more important one. Frictionless payments for AI agents are useful. Frictionless payments for AI agents that nobody authorized are a different kind of useful.
One number clarifies the stakes.
The most-used open protocol for AI agent payments has processed about $50 million in total transactions since it launched. The average transaction is thirty cents.
Mastercard processes roughly $9.7 trillion a year.
The agent payments market isn't being validated by Google and Mastercard and Visa moving into it. It's being absorbed. They moved at exactly the right moment: after independent developers proved the concept was real, before any of them had enough volume to have negotiating power. The idea gets credit. The infrastructure gets owned.
The governing body of the protocol that was supposed to be the open alternative to all this, the one that was built specifically so AI agents could pay for things without going through a commercial middleman, now counts Visa, Mastercard, Google, Amazon, and American Express as founding members.
That's not a criticism of the people who built it. It's just an observation about how capture works. The standard gets created by idealists. The foundation gets joined by incumbents. The incumbents bring transaction volume. The transaction volume becomes leverage. The leverage shapes the roadmap. One day you look up and the open alternative routes through the same three payment networks it was supposed to route around.
Whether that's adoption or capture depends on which direction the power actually flows. Right now there is one direction.
The thing to watch
The protocol that wins won't be the one with the best technology. It'll be the one that gets bundled into the tools developers already use. Whoever ships payment support inside their AI builder toolkit first, before anyone has to go looking for it, wins by default.
The press conferences already happened. The votes are the product decisions happening quietly in the background right now.