Because software cannot open a bank account, but it can control a crypto wallet and pay in seconds, at any hour, in amounts of a few cents. A May 2026 report from a crypto investment firm counted about $73 million settled by AI agents across roughly 176 million transactions in a year. It found 76% of those payments were smaller than the 30-cent fixed fee card networks charge.
A note on the legal and regulatory points here: they summarise public sources as of November 2026 and should not be relied on. Rules for AI agents are still being written, and some of what is described here is proposed or coming soon rather than settled. Check the current position, or ask a qualified professional, before acting on it.
Why can’t an AI agent just use a bank account?
An agent cannot open a bank account because the law asks for things software does not have. Under the US customer identification rule, a bank must collect, before opening an account, a name, a date of birth, an address and an identification number for an individual. For a company or trust, it needs a name, a physical place of business and an identification number, such as a taxpayer number.
An AI agent has none of these. It has no birthday, no street address and no tax number. It is not a legal person, so the bank has no one to identify.
The usual workaround is to let the agent borrow a human’s account. That works, but it means the agent is acting with your credentials, under your liability, through systems designed for a person clicking buttons. Our guide on whether an AI agent can own money explains why that ownership question never really goes away.
Even with a borrowed account, the economics do not fit. Card payments carry a fixed fee per transaction, so a payment of a few cents can cost more to process than it is worth. Agents tend to make exactly those payments, many times an hour.
What does a crypto wallet give an agent that a bank can’t?
A wallet gives an agent direct control of funds with no application, no opening hours and no human in the loop. Control comes from a private key, a secret number that signs transactions. The SEC’s investor bulletin on custody puts it plainly: the private key is what authorises transactions, and whoever holds it controls the assets.
Software can hold a key. That single fact explains most of the overlap between agents and crypto.
Wallets can also be programmable. An Ethereum standard for smart accounts, ERC-4337, lets each account run its own rules for approving a transaction, such as requiring several signatures or a custom recovery method. For agents, that opens the door to wallets with built-in limits, which later episodes cover in detail.
| Bank account | Crypto wallet | |
|---|---|---|
| Who can open one | A person or a registered company | Anyone, or any software, that can generate a key |
| Paperwork | Identity checks before opening | None to create; services may still check identity |
| Hours | Settlement follows banking days | Runs 24 hours a day, seven days a week |
| Smallest practical payment | Limited by fixed per-payment fees | Fractions of a cent on some networks |
| Rules on spending | Set by the bank | Can be written into the wallet itself |
Why do tiny payments matter so much?
Tiny payments matter because they are what agents mostly make, and card rails cannot carry them profitably. An agent might pay a fraction of a cent to read one page, call one data service or run one query. Multiply that by thousands of requests an hour and you have a business model that cards were never built for.
The numbers from the May 2026 crypto investment firm report show the shape of it. Over 12 months from May 2025, it counted:
- about $73 million settled by AI agents;
- roughly 176 million transactions;
- 76% of transactions below the 30-cent fixed-fee floor charged by card networks;
- 98.6% of payments settled in a single dollar stablecoin, a token designed to hold a steady $1 value.
The same reporting noted that a stablecoin transfer on one large layer-2 network, a cheaper network that settles back to Ethereum, costs about one hundredth of a cent. At that cost, a payment of a few cents is worth sending.
How does an AI agent actually pay for something?
Increasingly, through a revived web standard. The web’s HTTP rules have included a status code called 402 Payment Required since the 1997 version of HTTP/1.1, but it was never given a defined use. The current HTTP standard, from 2022, still describes it as reserved for future use.
An open standard called x402 now uses that code so agents can pay per request. In simplified form:
- The agent asks a server for a resource, such as a data feed or an article.
- The server replies with code 402 and its payment terms: price, token and network.
- The agent signs a payment from its wallet and resends the request with that payment attached.
- The payment is verified and settled, and the server returns the resource.
No account, no sign-up form and no card number are involved. Because the payment rides inside an ordinary web request, a service can charge per use rather than per month. That is the model agents need: pay for exactly the call you make, then stop.
The x402 design principles say it is network-, token- and currency-agnostic, and that the parties in the middle must not be able to move the payer’s funds without the payer’s say-so.
The wider internet is starting to take note. In August 2026 a draft proposal to define how the 402 code should work was published for discussion, citing emerging experimental uses. It is a draft, not an adopted standard.
What could go wrong?
The honest answer: crypto solves the payment problem for agents, but it creates new ones, and the market is still tiny.
It is small. $73 million over a year is tiny next to card networks. The trend is real; the scale is early, and most of the data comes from a handful of industry reports rather than official statistics. Treat claims of an “agent economy” worth trillions as forecasts, not facts.
It is concentrated. If 98.6% of agent payments run through one stablecoin, a problem with that one issuer, or one network it relies on, hits almost everything at once.
A key is a single point of failure. If software holds a key, anyone who compromises that software holds it too. The SEC bulletin warns that a lost private key means permanently losing access to the assets in that wallet.
Payments are final. Crypto payments typically cannot be reversed. There is no chargeback when an agent pays the wrong party, pays twice or pays a fake service.
Compliance does not disappear. A wallet needs no paperwork to create, but exchanges, stablecoin issuers and payment services that agents rely on still have identity and anti-money-laundering obligations. The legal questions move to the edges; they do not vanish.
The trust takeaway
Agents use crypto because it is the only money they can hold directly and spend in very small amounts. That same directness means no bank stands between the agent and a mistake. Before you fund any agent’s wallet, check:
- Whose wallet is it? Who holds the private key, and where is it stored?
- How much is in it? Fund it with only what the agent needs for the task, not your savings.
- What are the limits? Are spending caps or approved-recipient lists written into the wallet itself?
- What does it pay with? Know which stablecoin and network it uses, and what happens if either has a problem.
- Is there a record? Can you see every payment the agent made, to whom and why?