What are AI agents doing in crypto?

AI agents no longer just answer questions: given a goal, they browse the web, write code, use apps and carry out multi-step tasks on their own. Crypto gives them a natural layer for payments and ownership. This overlap is the most concrete part of “Crypto × AI”.

Why crypto?

An AI agent cannot open a bank account or get a credit card. A crypto wallet, however, can be created in seconds and can pay tiny amounts from anywhere online, around the clock. Dollar-pegged stablecoins make those payments possible without price swings.

What do agents do on-chain?

• Pay: settle per-use fees for a data service, an API or content.
• Manage assets: swap, provide liquidity or rebalance within the rules they are given.
• Sell services: sell analysis, content or compute to other agents and to people.

x402: the web’s “payment required” code

x402, proposed by Coinbase, puts the web’s mostly unused “402 Payment Required” status code to work. When a request hits a paid resource, the server answers 402 with payment instructions; the agent signs a stablecoin payment, retries the request and gets the data. No account or login is needed, and payment settles on-chain in seconds. Coinbase and Cloudflare announced an x402 Foundation to spread the standard.

This guide is for information only and is not investment advice for any token or project mentioned.

Numbers and risks

Adoption. According to figures compiled on Wikipedia, the x402 Foundation was formalized under the Linux Foundation in April 2026. Coinbase reported about 69,000 active agents, 165 million transactions and roughly $50 million in cumulative volume on x402 by late April 2026. Many transactions with a low average size show that most of these are small pay-per-use payments.

Agent tokens. Some projects tokenize their agents and tie the token to the agent’s revenue or governance. This area is highly speculative; prices often move on expectations more than on the agent’s real usage.

Main risks:

• Prompt injection: instructions hidden in a web page or message the agent reads can trick it into an unwanted payment.
• Key security: whoever reaches the agent’s wallet keys reaches its money. Spending limits and separate wallets are essential.
• Wrong decisions: AI models can be confidently wrong, and on irreversible on-chain transactions that means direct losses.
• Legal uncertainty: in many countries it is not yet clear who is responsible for a transaction an agent makes.

Our take

For now the most realistic Crypto × AI use case is “agents making small payments to each other”. Projects that focus on real transaction volume and security, rather than big promises, look like the ones that will last.