Agent Registry Convergence — Agent 365: Registry and Access

An agent registry is not just a list. In a large organization, it becomes the meeting point between visibility and control. One system must be good at noticing every artificial intelligence (AI) agent that exists, including the ones no one formally registered. Another must be good at giving each agent a durable identity, permissions, ownership, and protection. Registry convergence is the hand-off between those jobs: a discovered agent becomes a governed agent.

The problem is not the registry; it is the gap

Most governance failures begin in a gap between two truths. The organization has agents running, but the identity system does not know enough about them to govern them. Or the identity system is ready to manage agents, but only for the agents that arrived through approved channels.

That gap is where shadow agents live. A shadow agent is an unregistered or unapproved agent that may still call tools, read data, act on behalf of a user, or automate business work. It is not necessarily malicious. It may have been created by a team trying to move quickly. But if it is not visible and does not have a managed identity of its own, it is difficult to audit, restrict, retire, or hold accountable.

Agent registry convergence addresses this by connecting discovery to identity. Discovery answers the question, who is here? Identity answers, what is this agent allowed to do, who owns it, and how is it controlled? The registry is the place where those answers meet.

📌

Core idea. Discovery without identity creates awareness without control; identity without discovery creates control only for the agents you already know about.

Two jobs, deliberately separated

Microsoft splits the work between two systems with different strengths. Microsoft Agent 365 is built for breadth: discovering and inventorying agents across the organization and across platforms. Microsoft Entra Agent ID, where ID means identity, is built for depth: managing agents at scale with identities, access, ownership, lifecycle, and protections.

This split is an example of separation of concerns, a design principle that keeps different responsibilities in focused components. Finding every agent is a different problem from governing each agent well. The first rewards wide coverage and continuous inventory. The second rewards precise identity, policy, and access control.

The two systems are therefore complementary, not redundant. Agent 365 is the wide net. Entra Agent ID is the governed credential and control foundation. Convergence is the join between them.

Agent 365 provides breadth of discovery; Entra Agent ID provides depth of control.

What Agent 365 contributes: discovery and inventory

In this architecture, Microsoft Agent 365 provides the visibility side of the registry. Its role is to discover agents and maintain an inventory of them. Discovery means finding agents wherever they appear, including agents that were not intentionally registered and agents built or hosted outside the most obvious platform boundary.

Inventory means more than a one-time scan. A useful registry must be a running view of the agent estate: what exists, where it came from, how it is changing, and which agents need attention. This is why breadth matters. If discovery covers only approved build tools or only agents created by one central team, it will miss the very agents governance most needs to find.

Agent 365 therefore reduces the blind spots around agent sprawl. It is optimized for the first half of the governance challenge: make the agent population visible enough that it can be acted on.

Find agents across the organization rather than only in one sanctioned workspace.
Maintain an inventory that can be used by governance, security, and operations teams.
Expose agents that may not yet have clear ownership, identity, or policy treatment.

What Entra Agent ID contributes: identity and control

Microsoft Entra Agent ID provides the identity and access management foundation for agents. Identity and access management (IAM) is the discipline of proving what an actor is and controlling what that actor may reach. For human workers, IAM covers accounts, authentication, permissions, conditional access, and lifecycle events such as joining, changing roles, and leaving. Agents need the same kind of discipline, but adapted to non-human actors.

A first-class agent identity means the agent is not merely a label in a catalog and not a hidden script borrowing a user account. It has an identity that controls can attach to. Permissions can be granted to the agent. Ownership can be assigned. Activity can be audited against the agent. Lifecycle management can retire or change the agent without guessing which human or application credential it used.

This depth becomes more important as agents gain autonomy. The more an agent can call tools, use data, trigger workflows, or act over time, the more risky it becomes to treat it as an anonymous process or as a side effect of a user session.

Convergence turns found into governed

Registry convergence is the operating hand-off between Agent 365 and Entra Agent ID. Microsoft describes the mechanism plainly: agents discovered through Agent 365 can be assigned a first-class identity in Entra Agent ID. That hand-off is the important part. The agent does not remain only an observed object in an inventory. It can become an identity-bearing actor under the organization’s control model.

Once an agent has that identity, consistent governance becomes possible. Authentication can be tied to the agent rather than inferred from some surrounding context. Authorization can be expressed as what that agent may do. Lifecycle can be managed as the agent is created, changed, suspended, or retired. Policy can be applied in a repeatable way instead of handled as a one-off exception.

The cleanest way to think about convergence is as a pipeline. Agent 365 discovers and inventories. The registry provides the common view. Entra Agent ID assigns and manages the first-class identity. Governance then has something concrete to enforce against.

Convergence hands a discovered agent off to a governed, identity-bearing actor.

Why this closes the shadow-agent gap

A shadow agent can persist when visibility and control are separate states. If the organization sees an agent but cannot attach identity, access, and lifecycle controls to it, the agent remains known but weakly governed. If the organization can govern only agents that were intentionally enrolled, it misses the agents created outside the approved path.

Convergence reduces both failure modes. Discovery makes it harder for agents to remain unknown. Identity assignment makes it harder for discovered agents to remain unmanaged. The goal is not merely to build a better catalog. The goal is to shorten the distance between detection and control until found and governable are effectively part of the same process.

This is especially important for agents because they can combine software-like scale with user-like access patterns. An agent may perform repeated actions, call external tools, and interact with sensitive data. If it lacks a first-class identity, teams may struggle to answer basic questions: which agent did this, under whose ownership, with which permission, and why was it allowed?

✅

Governance gain. The value of convergence is that it removes the long-lived middle state where an agent is visible but not yet governed.

An agent is fully managed only where discovery and identity control overlap.

The practical decisions this creates

For leaders and practitioners, registry convergence is not just a product feature. It changes how agent governance should be designed. Teams need to decide what happens when a new or previously unknown agent is discovered. Is it automatically assigned an identity? Is it placed in a review queue? Is access limited until ownership is confirmed? The right answer depends on risk, but the question must be explicit.

Security architects should also decide which attributes are required before an agent is considered production-ready. Common examples include a named owner, business purpose, data classification, allowed tools, permission scope, and retirement plan. These details become more enforceable when tied to a first-class identity rather than stored as informal documentation.

Platform teams should treat the registry as an operational control point, not only an inventory screen. The registry can help route unknown agents to review, surface agents missing owners, identify over-permissioned agents, and support audit evidence. Convergence makes those workflows realistic because it connects the act of finding an agent to the controls that can manage it.

Define the default path for newly discovered agents: approve, restrict, quarantine, or review.
Require ownership before broad permissions are granted.
Use least privilege so each agent receives only the access needed for its purpose.
Make retirement and access revocation part of the agent lifecycle, not an afterthought.

When an unknown agent appears, ownership decides how much access it gets.

A simple mental model

Think of Agent 365 as the system that makes the agent estate visible and Entra Agent ID as the system that makes each agent governable. One is a discovery and inventory capability. The other is an identity and access foundation. The registry becomes powerful when those capabilities are connected rather than treated as separate administrative chores.

The result is a cleaner operating model for agent governance. Build or buy agents wherever the organization permits, discover them broadly, record them consistently, assign first-class identities, and apply policy through those identities. That model does not eliminate every risk, but it removes one of the most dangerous assumptions: that only approved agents need governance because only approved agents exist.

Key takeaways

Microsoft Agent 365 and Microsoft Entra Agent ID serve different roles in the agent registry rather than duplicating each other.
Agent 365 provides breadth by discovering and inventorying agents across the organization and across platforms.
Entra Agent ID provides depth by giving agents first-class identities, permissions, ownership, lifecycle management, and protections.
Convergence is the hand-off in which a discovered agent can be assigned a first-class identity and brought under governance.
The main security value is closing the shadow-agent gap between agents that are merely visible and agents that are actually controlled.
Practitioners should define clear workflows for newly discovered agents, including ownership, access limits, review, and retirement.

Glossary

Agent registry: A system of record that shows which agents exist and supports the governance work needed to manage them.
Artificial intelligence: Artificial intelligence is software capability that performs tasks associated with reasoning, language, perception, prediction, or autonomous action.
Convergence: Convergence is the hand-off that connects agent discovery to identity-based governance.
Discovery: Discovery is the process of finding agents wherever they run, including agents that were not formally registered.
First-class identity: A first-class identity is a dedicated, governed identity for an agent that permissions, ownership, monitoring, and lifecycle controls can attach to.
Identity and access management: Identity and access management is the discipline of proving what an actor is and controlling what that actor is allowed to access.
Microsoft Agent 365: Microsoft Agent 365 is the Microsoft control-plane capability described here as discovering and inventorying agents for broad visibility.
Microsoft Entra Agent ID: Microsoft Entra Agent ID is the Microsoft identity foundation described here as managing agents at scale with identities, permissions, ownership, and protections.
Shadow agent: A shadow agent is an unregistered or unapproved agent that operates outside normal governance.