Agent inventory
Governanceagent register · agent catalogue
A living register of every agent in production: what it does, who owns it, which systems and data it reaches, its autonomy level, which evals cover it and when it was last reviewed. If it isn't in the inventory, it shouldn't be running.
The agent inventory is the table that answers the uncomfortable audit question: how many agents do you have and what can they touch? Each row carries a business owner, purpose, connected systems, effective permissions, autonomy level, linked evals, monthly cost and last review date. This definition is Implementa's own editorial formulation. Inventorying AI systems is established practice in management frameworks such as ISO/IEC 42001 and in AI Act obligations, but the specific record we propose — effective permissions plus autonomy level plus evals on the same row — is ours, not an industry standard. Without an inventory, governance is a written policy nobody can verify.
How it differs from
- CMDB
- Inventories technical assets; carries no autonomy level, evals or business owner.
- Model registry
- Catalogues models and versions; the inventory catalogues agents in production and what they are allowed to do.
Examples
- A sheet listing an insurer's 14 live agents, their owners and the systems they touch.
- A quarterly review that switches off the 3 agents unused for 90 days.
FAQ
- Who maintains it?
- The client's AI operations owner. The technical team supplies the data, but ownership sits with the business.
- How often is it reviewed?
- Quarterly at minimum, and whenever an agent changes autonomy level.