You ask, we answer · 10 September 2026
What does ‘prepare our org for agents’ mean for IAM?
Our CISO told IAM to ‘prepare the organisation for agents’. We’re already drowning in service principals and guest accounts. Is this just another app-registration cleanup — or a real identity programme with Entra Agent ID and Agent 365?
Grounded with: Microsoft Learn — Overview of Microsoft Agent 365; Protect agent identities with Microsoft Entra; Agent Registry convergence; Governing Agent Identities
DIGITAL SKILL WORKS LTD / 12 PANES
You’re right to push back on a slogan. “Prepare for agents” without an owner, an inventory, and an identity model is how you get another year of mystery service principals — only now they chat.
Here’s what’s going on. Microsoft’s answer has two halves that must not be confused. Microsoft Agent 365 (generally available for commercial from 1 May 2026, per Learn) is the control plane to observe, govern, and secure agents across the estate — registry, lifecycle, and admin experience in the Microsoft 365 admin centre. Microsoft Entra Agent ID is the identity foundation: purpose-built agent identities (and blueprints) so agents can be authenticated, authorised, sponsored, and retired like other identities — not anonymous scripts wearing a shared secret. Learn’s registry convergence note is clear: Agent 365 for unified inventory and ops; Entra for Agent ID, Conditional Access, and identity governance signals.
So IAM’s job is not “build chatbots.” It’s to make every agent a first-class identity with an owner/sponsor, least privilege, and a kill switch.
Practical options:
- Wave 0 identity programme (recommended). Inventory what’s already running (Copilot Studio, Agent Builder, SharePoint agents, Foundry, anything with a classic app registration). Decide which need Entra Agent ID vs remain classic. Assign sponsors, apply Conditional Access / Identity Protection patterns you’d use for privileged apps, and wire lifecycle workflows so leavers don’t leave orphaned agents. Align with Agent 365 registry so security and M365 admins see the same truth.
- Registry-only theatre. Export a list from Agents → All agents, present a slide, change nothing about credentials or access. Looks busy; fails the first audit.
- Block everything and declare victory. User access = No users buys time; it doesn’t prepare you for the agents leadership already funded.
- Treat agents as “just another app registration.” You’ll miss blueprints, sponsorship, and Agent 365-specific lifecycle — and you’ll keep discovering shadow agents late.
The caution: Viewing inventory often needs roles such as AI Reader / AI Administrator; managing Agent IDs needs Agent ID Administrator (and related Entra rights). Full Agent 365 capability is licensed (Learn: at least one qualifying Agent 365 licence to enable; works best with E5 prerequisites — check current Microsoft plans, don’t invent SKUs). Agents that act on behalf of a user inherit that user’s access — Third Law still applies: convenience for the human is convenience for the agent. Autonomous agents need their own governed identity, not a borrowed break-glass account.
I’d suggest IAM returns a two-page Steady plan: (1) observe — registry ownership and weekly ownerless review; (2) govern — sponsor model and access packages for agent identities; (3) secure — Conditional Access, Purview/Defender hooks where licensed. That answers the CISO without promising a chatbot factory.
You’re asking the right question. Steady before scale means agents get identities before they get data.