I spotted a neat update from Microsoft this month that I reckon a lot of teams will find downright useful. Microsoft rolled out a human-in-the-loop feature for Copilot Studio at Ignite and added Agent 365 as a central control plane so organisations can build and govern agents more safely and practically for real work rather than experiments[3].
New Feature / Update: Human-in-the-loop and Agent 365 in Microsoft Copilot Studio
What is it?
In plain terms, Copilot Studio now lets AI agents pause and ask a real person for input mid-task, using structured review forms (for example via Outlook) so the agent can continue once someone approves or fills in the missing details[3]. It also introduces Agent 365, which centralises governance, policy, monitoring and auditing for all your organisation’s agents, so IT and admins can see what agents are doing and apply rules across the board[3].
Why does it matter?
This is not about flashy demos. It’s about making agents useful and safe for everyday work. Two simple, real-world ways teams might use it:
- Marketing campaign briefs: An agent drafts a campaign brief from product specs and previous briefs, then pauses to ask a marketing manager to confirm the target audience and budget via a short Outlook form. Once approved the agent populates the campaign in the project tracker and schedules creative tasks[3].
- Sales ops and CRM updates: An agent summarises a call transcript and proposes CRM field updates. It sends a one-question verification to the sales rep asking if the lead status should change. The rep taps yes or no in the Outlook form and the agent proceeds to update Salesforce and create follow-up tasks if approved[3].
Key practical details you’ll want to know
- Human-in-the-loop arrives as a preview capability and uses structured requests delivered as Outlook forms, so reviewers answer within familiar tools[3].
- Agent 365 centralises governance, audit logs and policy controls, helping IT enforce compliance and monitor agent activity across an organisation[3].
- Real-time protection is integrated with Microsoft Defender and identity control uses Microsoft Entra Agent ID so admins can put sensible guardrails on what agents are allowed to do[3].
How I’d try this tomorrow (short checklist)
- Pick one repetitive cross-team task such as drafting social posts from product notes or summarising weekly client calls.
- Create a simple agent in Copilot Studio that prepares the draft and is configured to pause for a single confirmation field sent to an approver’s Outlook.
- Enable Agent 365 governance settings so IT can log actions and set a limit on agent changes to CRM or billing systems.
- Run a week-long pilot with a small team and note time saved, number of pauses, and any noisy prompts that need trimming.
What I like about this update is how grounded it is. It does not promise agents will do everything; it gives teams a way to make them part of a safe workflow. Also, a tiny detail that made me smile: one of the example Outlook forms had a cheeky single-question prompt that read, “Change lead to opportunity? Yes / No / Ask me later.” Too real to be fake, that one, and exactly the kind of small UX thing that stops approvals turning into email marathons[3].
Sources: Microsoft Copilot Studio November 2025 updates and Ignite announcements[3].
