Your AI just made several payments while you were reading that headline. You approved none of them. Visa processed none of them. And if the crypto industry’s biggest bulls are right, it’s not a mistake – it’s the entire future of the internet economy.
Coinbase founder Brian Armstrong believes that soon there will be more AI agents than humans conducting transactions on the Internet. Binance founder Changpeng Zhao went further and predicted that agents will make a million times more payments than people, all in crypto. The posts landed on the same day last week and lit up crypto X.
The core argument is structural.
AI agents cannot open bank accounts because banks require identity verification that software cannot provide, while a crypto wallet only needs a private key. No KYC, no compliance review, no waiting time – and that asymmetry is what Armstrong pointed to.
But the wallet problem is only half the picture. The other half is economics.
Agents do not act as humans do. When an AI agent performs a task—such as researching a topic, coordinating a supply chain, building a report—it can call dozens of specialized APIs in a single session.
Each call can be worth fractions of a cent, paying for GPU compute time, real-time data feeds, web scraping services, or hiring a sub-agent to handle translation. None of these transactions resemble anything that Visa or Mastercard are designed to process.
Consider for a moment that this story was written by an agent, requested by a “boss” agent at CoinDesk tasked with increasing the site’s authority.
To produce it, the agent would have queried a real-time news API to verify Armstrong’s tweet ($0.002), pulled onchain data to search for volume figures ($0.004), cross-referenced press releases ($0.001), and pinged a financial context model for Visa protocol details ($0.003). It would finally generate the article for an additional cost, paying credits to another AI tool for actually write the piece.
The total cost of reporting is under two cents with six transactions at the current numbers offered by protocols such as x402.
In contrast, Stripe’s minimum processing fee on a single transaction is around $0.30. Running these six payments through a card network would cost more than 100 times the value of the payments themselves.
A human editor who reviews and publishes the piece might then be billed by a sub-agent who handled SEO optimization, another who ran plagiarism checks, and another who formatted for CMS software. Each micropayment is economically absurd on card rails, but trivial on the chain.
This is the thesis behind x402, Coinbase’s open payment protocol that integrates stablecoin payments directly into HTTP requests – so an agent can hit a paywall, pay in USDC and continue their task in the same interaction, without the need for humans. Cloudflare, Circle, AWS and Stripe all back it. Google’s open agent payment standard includes x402 as a settlement layer.
Any industry with high-frequency, low-value data exchange becomes a candidate.
In healthcare, an agent who administers a patient’s insurance claim pays per document retrieved from a medical record API. In logistics, a purchasing agent auctions cargo slots across dozens of carriers in real time and determines the winning bid instantly. In the media, AI crawlers pay per article indexed instead of negotiating bulk license agreements. In finance, a trading agent pays a specialist model fractions of a cent per consumed risk signal.
One caveat, however, is that infrastructure is ahead of demand.
CoinDesk reported this week that x402 currently processes about $28,000 in daily volume, with Artemis marking about half of the observed transactions as artificial activity rather than real trading. The merchants the x402 was built to serve are still rare.
Meanwhile, traditional finance is not standing still. Visa launched its Trusted Agent Protocol last October and Mastercard completed Europe’s first live AI agent bank payment inside Santander’s regulated infrastructure last week – both on existing card rails with cryptographic verification superimposed.
The most likely outcome is a split where regulated trading remains on card rails while machine-to-machine payments – such as agents hiring agents pay per API calls, buying compute on demand – migrating to stablecoins because the economy demands it.
The open question is which bucket ends up bigger.
