DefinedTerm · agents

Agent execution protocol

The rules under which an AI agent is permitted to do consequential work, made machine-checkable: who authorized the agent, what it may act on, which decisions escalate to a human, and how finished work is sealed into a record a counterparty can recompute. It matters because without a protocol every one of those answers is a convention in someone’s head, and conventions do not survive a fleet.

Common causes

Capabilities that name departments instead of actions, so nothing can be enforced; uniform approval gates everywhere, so governance becomes theatre people route around; work that completes with nothing sealed.

What good looks like

A manifest naming an accountable human per agent; capabilities as verbs; approval thresholds only where a mistake would hurt; deterministic repository routing that never guesses; settlement payloads frozen as canonical JSON plus sha256 at resolution.

Used in

Nearby terms

This is the build-lens entry, deliberately narrower than the market framing of the same subject: see What is an agent execution protocol? — FlashyOS.

Term set: the build vocabulary. Capital-markets terms are defined on gda.group; agent-economy terms on flashygroup.com. One term, one home.