Skip to content

AI agents and automation · News explained

Microsoft Agent 365 at Ignite 2026: who controls your AI agents?

Microsoft’s Agent 365 updates focus on visibility, identity and cost. Learn the difference between building an agent and governing what it can do.

Jump to practical guide
Concept illustration of a person reviewing a network of governed AI agents
AI-generated concept illustration. Not an event photo or product screenshot.

Microsoft’s Agent 365 updates focus on visibility, identity and cost. Learn the difference between building an agent and governing what it can do.

What changed at Ignite

On 17 November 2026, Microsoft announced expanded Agent 365 capabilities for observing and governing Microsoft and non-Microsoft agents. The update includes Business Analytics in Frontier preview, additional partner identity connections in preview, dynamic risk views and multi-tenant management in public preview.

Microsoft — Agent 365 Ignite 2026 update

Why preview labels matter

Microsoft describes an administrative view of agent ownership, access, risk and spending, with entitlements varying by capability. A preview announcement does not mean every tenant can immediately use every feature. Its getting-started route begins in the Microsoft 365 admin centre.

Microsoft — Agent 365 Ignite 2026 update

What it means for learners

An agent can read information or take actions through tools. Governing it means knowing who owns it, what it may access and how to stop it. Our practical takeaway for a small team is to create this inventory before buying more automation. A spreadsheet with clear ownership can expose problems that another chatbot interface will not solve.

Picmkr practice exercise

Try a small, reviewable project

What you need A list of your team’s experimental agents. An eligible Microsoft tenant and administrator are needed only if you want to evaluate Agent 365 itself.

  1. Create an agent register

    For each agent, record its purpose, owner, platform, model, tools, data sources and where it runs. Include unfinished experiments, not only production systems.

  2. Describe allowed actions

    Separate reading a course sheet from changing a booking or sending an email. Record the exact approval point for each action with an external effect.

  3. Check tenant availability

    If your organisation uses Microsoft 365, ask the authorised administrator to follow the official setup guide and confirm licensing and preview eligibility. Do not change tenant-wide settings for a personal test.

  4. Test suspension and review

    Use a harmless test agent and fictional data. Verify that its owner can inspect its activity and disable its access. Keep a record of the actual result.

Sample brief

Agent register fields: name, purpose, human owner, platform, model, readable data, writable data, allowed tools, approval rule, spending limit, review date and disable procedure.

How to tell it worked

Every agent should have a human owner, a defined scope and a tested way to stop activity. Record unavailable preview features as unavailable.

Governance visibility does not itself prove that a workflow is correct. Continue testing the underlying task and checking approvals.

Common question

Do I need Agent 365 to learn AI agent governance?

No. You can practise ownership, least-privilege access, review logs and approval rules in a small manual inventory.

Sources and further reading

Announcement dates are shown above. Vendor performance claims are attributed; Picmkr has not independently benchmarked these releases. Practical exercises are editorial suggestions.