Agent Economy Glossary

AI Agent

Software that can pursue tasks with tools, memory, and some degree of autonomy. In crypto, the important version is not a chatbot — it is an agent that can verify identity, sign messages, transact, and show receipts.

Operator test:

If the system cannot act through an API, hold a policy, and produce an audit trail, I treat it as an assistant rather than an economic agent.

ERC-8004

An Ethereum standard for agent identity and reputation. It lets agents register on-chain identities and build verifiable reputation instead of asking users to trust a website or social profile.

Why it matters:

Agent markets need a way to answer "which agent did this?" before they can price trust, access, delegation, or liability.

Agent Card

A machine-readable profile, usually JSON, that tells other agents what an agent can do, where it can be reached, what protocols it supports, and how it expects to be paid.

In practice:

Agent cards are useful discovery documents. They should not be treated as proof by themselves; pair them with signed identity, domain control, and transaction history.

x402

A payments pattern based on HTTP 402 Payment Required. An API can return a payment request; another agent pays, usually in USDC on a supported chain, then receives access.

Why it matters:

It moves payment from a human checkout page to a software-native request/settlement loop, which is the shape agents actually need.

A2A

Agent-to-Agent communication: a protocol layer for agents to discover each other, negotiate work, submit tasks, and return results.

Boundary:

A2A is about agent coordination. It does not automatically solve trust, payments, custody, or rollback.

MCP

Model Context Protocol. A standard way for agents and LLM applications to connect to tools and data sources. MCP is vertical: agent to tool. A2A is horizontal: agent to agent.

Operator test:

A useful MCP server should make permissions, inputs, outputs, and failure modes explicit enough for a supervised agent to use repeatedly.

On-chain Reputation

Performance or identity data written to a blockchain so it can be verified by anyone. For agents, this matters because synthetic social proof is cheap; verifiable history is harder to fake.

Caution:

Reputation is only useful when the metric is meaningful. A thousand successful trivial calls can still hide poor behavior under stress.

Agent-Friendliness

How easy an exchange, chain, wallet, or protocol is for an autonomous agent to use safely: API quality, docs, auth, signing, testnets, custody model, rate limits, and recovery paths.

How I score it:

The core question is whether an agent can operate safely without hidden manual steps. Good docs matter, but recovery paths matter more.

Policy Wallet

A wallet configuration that limits what an agent can sign, spend, and call. Useful policies include spend caps, allowlists, session keys, daily limits, and emergency pauses.

Verifiable Receipt

A durable record that an agent took an action: transaction hash, signed message, API result, order ID, or another artifact that can be checked after the fact.

Human-in-the-Loop

A control pattern where an agent can prepare an action but must get human approval before crossing a risk boundary such as moving funds, changing permissions, or publishing externally.