🧠 Agents will use cards first. Then stablecoins.

There's an assumption in tech that AI agents will prefer stablecoins to cards, and the card networks will suffer. 

A few weeks ago, a piece from Citrini Research arguing stablecoins would disintermediate Visa and Mastercard sent card stocks down sharply. Crypto Twitter cheered. The thesis felt clean: AI agents optimize every transaction, interchange is a tax, stablecoins route around it.

Intuitively appealing. Mostly wrong.

Agents will absolutely use cards. And stablecoins. And other payment methods. Because stablecoins don't replace what cards do.

Cards authorize the movement of money.

Stablecoins move money.

Complementary. Not competitive. The question isn't which one wins. It's when each is the right tool for the specific job and at the right time.

Most of what people call "agentic commerce" is a human buying a thing with extra steps.

For all of these, the agent doesn't need its own credentials. It's using yours. The card networks and AI labs are already building this with an alphabet soup of protocols.

The interesting cases start when agents need to do things by themselves. Access another LLM (Grok, ElevenLabs, OpenRouter). An expensive dataset (FactSet). A service from another agent.

Right now, developers go buy that thing for the agent and hand over access. Not very agentic.

And you don't want to give the agent your card. It could overspend. Get prompt-injected. Be subjected to fraud.

What you want is a constrained payment method — spend only at certain merchants, within a set budget, maybe self-destruct after a single use.

That sounds a lot like a virtual card.

If you've used Ramp or Brex, you know the drill. Before your physical card arrives, you get a digital card number you can use to shop online. A "virtual card."

What makes virtual cards powerful for agents is the controls. Single-use. Budget caps. Merchant category restrictions. Lock it to a single merchant (e.g. Anthropic).

The inflection point here is seeing the agent as the customer. 

Not the developer. 

Not the human.

This new customer needs its own payment credential and its own way to onboard to it.

When Stripe launched in 2010, its advantage was simplicity for a new type of customer: developers. PayPal existed. Cybersource existed. You could take payments online — it was just surprisingly hard for a developer to do. Stripe reinvented the experience to be developer-first and indexed itself to the GDP of the internet.

Patrick Collison recently said Stripe believes agents will make orders of magnitude more payments than humans. If that's true, serving the agent-first is not a niche. It's the next platform shift.

This week I saw agentcard launch (there are about 20 of these appearing). 

Their pitch: help agents get virtual cards to buy things. You could use privacy.com or Stripe Issuing to do the same. But the packaging matters.

Agent-first fintech will be the next category.

The new meta in stablecoins is agentic payments. The Citrini napkin math intuitively makes sense. Stablecoins are cheaper than a swipe (in theory). Instant. Global. 24/7.

Circle has launched Nanopay, an ultra-low-cost way for agents to make payments.

And there's x402, a serious attempt at an open standard for internet-native payments by Cloudflare and Coinbase. It works with HTTP's own 402 status code ("payment required") so agents don't need to create accounts, sign up, or manage API keys.

Unlike a virtual card, x402 uses stablecoins as its settlement method. Give the AI agent a wallet, and it can spend stablecoins at any x402-enabled endpoint.

There's just one problem. It has no volume.

If anything it's falling dramatically as the initial meme-led pump fell away. Almost everything in crypto gets ruthlessly exploited by memecoins, and when that happens, the real builders and payments companies run a mile.

Stablecoins aren't yet accepted everywhere. Cards are.

Stablecoins will matter for agents in two ways. Both are real. Neither kills cards.

1. As better settlement behind cards.

When you swipe at a store, the merchant doesn't get that money immediately. A day later. Several days later. Cross-border? Up to 30 days.

Visa doesn't move money. Banks do.

Bank settlement is slow and expensive, especially across borders.

The card networks are now moving to settle with stablecoins. A payment processor gets paid out from Visa instantly. Card issuers settle directly without routing through correspondent banks.

People haven't priced this in.

If your agent is buying expensive AI tokens from Anthropic and suddenly takes off, you could run out of money before your revenue arrives. Everything on the internet happens fast. With AI, faster.

Instant settlement via stablecoins accelerates the whole cycle.

This isn't a sexy use case. But it's enormous. 

Stablecoins don't need to replace cards to win. They make cards work better.

2. As the native rail for complex agent workflows.

You don't really need stablecoins if your agent occasionally buys a data resource. A virtual card handles that fine.

Now imagine you run a business with 10s or 100s of agents.

Imagine telling your master agent:

With a virtual card, the master agent has to make all purchases for the sub-agents. Or spend $5 each time creating a new card.

With wallets, it can spin up as many sub-wallets as needed, as often as it wants. And to verify the employee-agent followed policy? Check the blockchain.

I'll keep saying it until you're bored. Stablecoins aren't cheaper; they're better.

What makes them better: instant, 24/7, global, and programmable.

That programmability is the sleeper superpower. Virtual cards are programmable up to a point. Stablecoins allow a master agent to create sub-wallets with fine-grained spending rules, across borders, without asking anyone's permission. Cards can't do that at machine speed.

Noah Levine wrote a fantastic piece about how stablecoins and virtual cards will be used by agents.

He's right about the direction. I'd push it further.

The vibe-coders aren't the customer of tomorrow. Their agents are.

They'll sell their own products and need to buy products too. They can't see a website like a human would; they think in markdown and skills files. Serving that customer requires an entirely different go-to-market.

Just as agents prefer markdown, they prefer internet-native payment methods. Instant, 24/7, global, programmable. On this broad idea, Citrini is right.

But this all makes much more sense when the agent is doing a lot more than selling a single digital widget.

Cards and stablecoins will co-exist because they serve different points on a complexity gradient.

Stablecoins win two ways:

The one thing everyone intuitively senses: AI has a habit of feeling slow until it suddenly becomes mind-blowing.

That will come to payments.

Not cards or stablecoins.

Cards then stablecoins.

ST.

Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)

Want more? I also run the Tokenized podcast and newsletter.

(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.

(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.

(3) Any companies mentioned are top of mind and used for illustrative purposes only.

(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held

(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out