01
Find every agent.
Sanctioned, shadow, embedded. Telemetry, not surveys. Personal ChatGPT seats and embedded copilots show up in the same view as your blessed deployments.
AI Register.
An employee record for every agent, model, and copilot.
When AI does the work, AI needs HR. The Register inventories every agent in your org, profiles what it does, who owns it, and whether it's earning its keep — same record format every time.
An HR system for non-human workers.
Three things every AI register must do, today, before the regulator's question lands in your inbox.
02
Profile each one like a person.
Scope of work, decisions per week, override rate, accuracy trend, peer comparison, human DRI. The agents that can't fill out their own profile are the ones to retire.
03
Govern at AI speed.
PII exposure, ownership gaps, model drift, regulatory class — surfaced before the incident, not during the post-mortem.
Inside a typical Register
147
agents found in week one — most customers expected ~40
38
personal ChatGPT seats discovered, on average, in mid-market customers
11%
of AI decisions overridden by humans — the early signal of drift
100%
of agents assigned a human DRI before sanction
A record for every agent. The same record every time.
The point of the Register is that you can compare agents the way you compare employees — same fields, same rubric, same accountability.
The profile
What an agent does, in language a manager understands.
Each profile carries the workflow it sits in, the data it reaches, the decisions it makes per week, its accuracy trend, and the human responsible for its work. The structure is the same whether the agent is a vendor copilot or a hand-rolled internal one.
- WorkflowLead qualification
- Decisions / week1,247
- Override rate11.2% (−2.1pp MoM)
- Data scopeCRM · email · public web
- Human DRIMaya P. · RevOps
The risk view
Risks the regulator will be looking for.
Built around EU AI Act risk classes, NIST AI RMF, and ISO 42001 vocabulary. Map each agent's risk class once; let the Register monitor it forever.
- ClassEU AI Act · limited risk
- PII exposureNone (synthetic IDs)
- Drift watchAccuracy delta vs prior month
- Owner gaps0 (was 4 at onboarding)
- Audit exportAuditor-ready, single click
The lifecycle
Onboard, evaluate, retire — like any other worker.
Sanctioning an agent is a structured event. Quarterly evaluations are scheduled. Retirement is logged against the work that absorbed the agent's responsibilities. The lifecycle is auditable.
- OnboardingRequired: DRI, scope, evals
- Eval cadenceQuarterly · with override review
- RetirementLogged · workflow reabsorption noted
- Audit trailFull lifecycle, exportable
If you wouldn't hire a human without a job description, don't deploy an agent without one.
Node — AI governance team
Stop guessing your software decisions.
Node maps how your business works and how software supports it — so every decision is grounded in what's real.
Book a demo