Keyrock has published a report on agentic payments, arguing that the next hard problem for autonomous AI agents is not just finding tools or calling APIs, but paying for services without constant human intervention.

The report, published May 21, says the market for agents spending money online is still small. CoinDesk, summarizing the report, said large technology, payments, and crypto companies are nevertheless already building infrastructure around the idea. The practical use cases include agents buying data, compute, identity checks, content access, or task-specific services on behalf of users.

Keyrock frames crypto rails as one candidate architecture because they can combine programmable wallets, stablecoin settlement, and machine-readable payment rules. That does not mean agent payments are ready for mass adoption. The open questions are still substantial: spending limits, revocation, fraud controls, compliance, identity, and how a user proves what an agent was authorized to buy.

The work also sits alongside broader standards efforts for agent economies. ERC-8004, for example, proposes registries for agent identity, reputation, and validation so agents can discover and evaluate one another across organizational boundaries. It is not a payments standard, but it points to the same direction: agents need shared infrastructure before they can reliably transact outside a single platform.

For now, the news is less that AI agents are already autonomous consumers and more that payments infrastructure is moving into the design phase. Stablecoins and onchain authorization are increasingly part of that conversation.