Agent Registry

What is an Agent Registry?

An agent registry is a central catalog of the AI agents running in an organization. For each agent it records the facts governance depends on: who owns it, which product or team it belongs to, what it is for, which tools it declares and which endpoints it talks to. People, programs and other agents query it to discover what already exists and to decide what an agent is allowed to do.

The registry answers questions that become impossible once agents spread across teams as chatbots, glued-together MCP servers, notebook experiments and internal web tools: who owns this, what can it touch, and what should it not be doing?

Why It Matters

• You can't govern what you can't see - Without an inventory, a company can't assess risk or respond to incidents. See Agent Sprawl.
• Reuse instead of duplication - Teams that can't find existing agents rebuild the same thing.
• Policy at runtime - At Two Sigma, an agent starting in a managed environment looks up its policy in the registry, and that policy becomes enforceable rules on endpoints, hosts and resources.
• Risk-based review - Declared tools and endpoints let a firm score each agent, so low-risk agents skip the review queue meant for agents that touch production.

Registration Without a Gate

The common mistake is making registration an approval gate that every agent waits behind. John Feminella of Two Sigma argues this pushes people to shadow IT, where nothing is visible. His alternative is automatic registration: building an agent with an approved framework registers it as a side effect, with manual registration available for everything else.

Optiver takes a related approach with an agent gateway that acts as a single ledger of every agent in the company, so agents and humans share identity management and agents can call each other.

Related Reading

• John Feminella: Two Sigma's 5 Controls for AI Agents - The registry as the first of five controls
• Noyan Tokgozoglu: How Optiver Ships New AI Models Same Day - An agent gateway as a company-wide ledger
• Agent Observability - Seeing what registered agents actually do
• Agent Sprawl - The problem registries solve