Skip to content
Implementa.
← Full glossary

Agent registry

Practice

agent inventory

A living inventory of every AI agent an organisation runs in production, with its owner, the process it operates, the systems it can reach, its autonomy level and the date of its last evaluation. This is an Implementa editorial formulation: we treat it as the first governance deliverable of any deployment.

The idea is mundane, which is why it works: you cannot govern what isn't on a list. A useful agent registry answers five questions per row — who is accountable for this agent, which business process it operates, which systems and data it reaches, how much autonomy it has (assisted, supervised, autonomous) and when it was last evaluated. Implementa treats it as a governance deliverable, not documentation: it's what lets you answer an audit, decide what gets switched off when a regulation changes, and catch agent sprawl before it becomes unmanageable. Without a registry, the first question in any audit — "how many agents do you have running?" — has no answer.

How it differs from

Agent observability
Observability tells you what an agent is doing right now. The registry tells you which agents exist, who owns them and with what permissions. One is telemetry; the other is a census.
Agent identity
Identity is the credential an agent acts with inside your systems. The registry is the organisational record tying that credential to a human owner and a process.

Examples

  • One row per agent with process owner, connected systems, autonomy level and last evaluation date
  • Flagging in the registry which agents touch personal data, so you know which ones a new regulatory obligation hits

FAQ

Is a spreadsheet good enough?
At the start, yes — and it beats nothing. The registry fails when it turns into documentation nobody updates. What keeps it alive is tying it to a mandatory moment: no agent reaches production without a row, and no row survives without a recent evaluation date.
How do you notice you don't have one?
You notice the day someone asks how many agents are running, who is accountable and what data they reach, and the answer takes three weeks. You also notice it on the bill: forgotten agents keep consuming.

Related terms