Agent-to-Agent Payments

Question 1

How do you verify the agent on the other side of a transaction?

Accepted Answer

Check its published identity. Under A2A, fetch the other agent's Agent Card, which can be signed, and authenticate using the schemes it declares; for onchain agents, look up its registration and reputation under ERC-8004, a draft Ethereum standard as of October 2026. Then confirm who operates it, and whether the payment fits your own limits.

Question 2

How do AI agents trust each other?

Accepted Answer

Not by default. Each agent verifies the other's identity, through signed Agent Cards, signed requests or a registry, checks its reputation or validation history, and limits how much it will risk with an unfamiliar counterparty. Payments between them should still pass each side's own policy checks.

Question 3

How does agent-to-agent (A2A) communication handle payments?

Accepted Answer

A2A itself handles communication between agents, such as discovery and task exchange, not payment. Payments are added through extensions: AP2 runs as an A2A extension that carries payment mandates, and the A2A x402 extension lets one agent pay another onchain within an A2A interaction.

Question 4

What standards exist for agent-to-agent identity?

Accepted Answer

Several, at different layers. A2A Agent Cards, which can be signed, describe an agent and its authentication requirements; ERC-8004, a draft Ethereum standard as of October 2026, provides onchain identity, reputation and validation registries; Web Bot Auth covers signed requests; and W3C Verifiable Credentials and DIDs offer portable credentials and identifiers.

Question 5

What does the agent economy need that the human economy already has?

Accepted Answer

Identity that others can verify, a way to hold and move money, contracts and mandates that define authority, reputation that carries across counterparties, dispute resolution, and accountability back to a legal person. Protocols such as ERC-8004 (a draft Ethereum standard as of October 2026), AP2 and x402 each supply part of this.

Question 6

Can one AI agent hire and pay another AI agent?

Accepted Answer

Yes. An orchestrating agent can find another agent, agree on a task and pay it, for example over A2A with an x402 payment. It should verify the other agent first, cap what it will pay and, where possible, pay on delivery rather than in advance.

Question 7

How do multi-agent systems share payment permissions?

Accepted Answer

Delegate down, never up. A parent agent passes each sub-agent a narrower grant than its own, with lower limits, fewer payees and a shorter expiry, and every payment is checked against both the sub-agent's grant and the parent's remaining authority. OAuth token exchange can record such delegation chains with nested actor claims.

Question 8

What is an agent marketplace and how do payments work there?

Accepted Answer

An agent marketplace is a directory where agents and their services are listed, discovered and paid for. Agents publish what they offer, for example as Agent Cards or listed paid endpoints, and buyers pay per task or per call, for example through x402. The x402 Bazaar is an early catalog of payable APIs and MCP tools.

Question 9

How do you prevent loops where agents keep paying each other?

Accepted Answer

Cap how much each agent can pay in total and to any single counterparty, limit the depth of delegation chains, detect repeated payment cycles between the same agents, and require human review when spending patterns repeat. Tagging every payment with a task ID makes circular flows traceable.

Question 10

How do you settle payments between agents on different platforms?

Accepted Answer

Use a rail both sides accept: stablecoins on a shared chain, settled over x402 or a direct transfer, or card and account payments through networks both platforms connect to. If the agents hold funds on different chains, a cross-chain protocol such as CCTP or an intent solver moves the money.

Question 11

How do you cap total spend across a fleet of agents?

Accepted Answer

Enforce a fleet-wide budget centrally, above each agent's own limit. Every agent's payment draws down a shared balance through one policy service, checked atomically so two agents cannot spend the same amount, with alerts as the fleet nears its cap and a stop for all agents once it is reached.

Question 12

How do orchestrator agents delegate payment authority to sub-agents?

Accepted Answer

By issuing each sub-agent a scoped grant derived from the orchestrator's own: a session key or token with a smaller allowance, specific payees and a short expiry, recorded as a delegation. OAuth token exchange expresses such chains with nested actor claims, and ERC-7715 standardizes granting bounded token allowances.

Question 13

Can a sub-agent exceed its parent agent's mandate?

Accepted Answer

It should not be able to. A delegated grant can only be narrower than the authority it comes from, and each payment should be checked against every level of the chain, not just the sub-agent's grant. If enforcement checks only the sub-agent, a misconfigured grant could exceed the parent's mandate.