Envisago
Artefact | Governance

The Envisago Agent Register

One page that holds every agent acting in your operation: what each one is, who owns it, where its authority stops, how quality is checked and what would retire it. Free, and complete in itself.

Download the register (Excel) Version 1.0, July 2026. No email address required.
Why a register

Agents arrive one at a time. Together they become hard to govern.

Adopted by different teams for different reasons, agents accumulate, and governing them becomes a job in itself: who owns what, where authority stops and how quality is checked gets harder to answer with each new arrival. The register is how the operation keeps up: every agent, its classification, its owner, its limits and its evidence, on one page the whole operation can see. Most of its value appears the first time you fill it in.

Empty owner boxes, empty never-allowed boxes and empty retirement conditions are not gaps in the form. They are the current state of agent governance in your operation, made visible.

What belongs on it

Four kinds of agent are in scope.

Workflow agentOwns one step or segment of a defined workflow. Its output flows to the next step in the process.
End-to-end agentHandles a whole task from start to finish, dealing directly with customers, suppliers or systems, with no person in the loop per case.
Ambient agentPresent all the time in a shared space such as a team chat channel, answering whoever asks. Shared and always on, so no one person owns each output.
OrchestratorCoordinates other agents or sequences a pipeline: it decides what runs, in what order and where exceptions go.

Not on the register: personal assistants, the AI a person works with one to one inside their own task, and the sub-agents those tools spawn. The person reviews that output before it goes anywhere, so the person is the control.

The test for any borderline case: if its output reaches the operation without the person who started it owning each result, it goes on the register.

What the register holds

Six sections. Every column surfaces a finding.

IdentityClassification, function, scope (single function or shared), what it does, the workflow it sits in, what it is built on and the IT record it links to.
OwnershipOne named owner who answers for the agent, and its day-to-day supervisor.
AuthorityWhat it acts on alone, what needs a person's approval and what it is never allowed to do.
QualityHow output quality is checked, by whom, how often, and where it sends what it cannot handle.
ReviewThe review cadence and the small set of measures actually watched.
LifecycleStatus, live date and the retirement condition, written before you need it, with the holder of the off switch.

The download includes four worked examples for a fictional operation, one per classification, and the method for filling it in: list every agent in scope, give each one an owner, draw the authority line, set the quality check, set the review and write the retirement condition before you need it.

Download the register (Excel) This page always holds the current version.

The Agent Register is part of AI Operating Model Design and sits within the Design dimension of AIVOM™, the AI Value Operating Model. For a structured read of where your operation stands across all four dimensions, start with the free AI Operating Impact Briefing.