Why an AI Workforce Needs a Constitution

Most organizations meet agent governance as a single problem, usually during an incident. It is three, and they are built in order: an agent must be identifiable before its authority can be scoped, and both must be recorded before anything can be audited. Skip the first and the other two are decoration.

Three disciplines, three homes

Identity — who the agent is and what it may be asked for. The operator account is at FlashyOS; the liability question at GDA Group.

Permissions — what it may commit alone. Operator account at FlashyOS; the board framing at GDA Group.

Audit — what it did and why that was allowed. Operator account at FlashyOS; what an auditor actually asks for at GDA Group.

Why a constitution rather than a policy

A policy document describes intent and is read once. A constitution is enforced by the system that runs the organization — the approval gate that will not open, the credential that was never issued, the record that cannot be edited. The distinction is the whole argument: governance that lives in a document is a hope, and governance that lives in the mesh is a control.

Flashy Group runs this way across ten properties on one mesh, which is why a new property inherits the rulebook rather than drafting one.

Governance is the half of the agent economy that decides whether the rest is safe to run. The Agent Economy covers it as part of the assessed track.