Almanac
Microsoft/copilotMicrosoft 365

Consultant KB for the whole Microsoft Copilot estate. Spans Microsoft 365 Copilot, Copilot Studio, the Dynamics 365 Copilots, Power Platform Copilot, Fabric & Power BI Copilot, Copilot in Azure, Windows & consumer Copilot, extensibility & agents, Security Copilot, and GitHub Copilot, plus admin/security and cross-product decision guides. Populated by the daily author agent from Copilot release notes / release plans plus the author's own consultant notes.

feature-agentic-governance-power-platform-admin-cowork.mdv1 · history
CurrentApplies to M365 Copilot (Cowork), Power Platform adminUpdated 2 weeks agoSource Microsoft Learn ↗

Status: Public Preview — behaviour may change.

What it does

Microsoft's adoption guidance for running tenant governance as a conversation rather than a reporting pipeline. You ask the Power Platform Admin plugin a question in Cowork, such as which apps nobody has opened in 30 days, then ask Cowork to turn the answer into a report, draft an email to the owners, or repeat the question on a schedule. The plugin supplies the tenant knowledge; Cowork supplies the follow-through.

Key facts

  • The plugin is read-only. It gathers inventory, usage and ownership data across environments and doesn't change settings or take admin actions itself.
  • Intended users are Power Platform administrators, Dynamics 365 administrators and Global administrators.
  • Questions consume Copilot Credits. Spend is monitored and capped through spending policies in the Microsoft 365 admin center, not in Power Platform.
  • Managed Environments aren't needed to ask questions. Features you then act on, such as environment groups and default environment routing, keep their own requirements.
  • Cowork asks for your approval before sensitive actions such as sending an email or posting in Teams, and prompts can be scheduled from the Automations view.
  • The guidance comes as a series: an overview, a prompt library (unused and ownerless resources, default environment clean-up, connector-block impact, showback by department, security baseline, capacity forecasting) and a page on turning answers into HTML reports, scheduled briefings and reusable skills.
  • Microsoft states plainly that answers are AI-generated and vary between runs and tenants.

When to use / skip

Use it to get a first-pass picture of an unfamiliar tenant, or to replace the weekly "what changed" spreadsheet someone maintains by hand. The connector-impact prompt before a data policy change is the most practical case. Skip it as the evidence base for anything contractual or audit-facing; it's preview, probabilistic and read-only.

Configuration decisions

  • Run a month-long pilot of two or three scheduled scenarios and size the Copilot Credit budget from actual consumption, which is Microsoft's own suggestion.
  • Set the spending policy before the rollout, not after the first invoice-shaped surprise.
  • Decide who receives the scheduled briefings and whether owner outreach stays as drafts (the prompt library suggests saving to Drafts) or is sent after approval.
  • Agree with compliance on regions and AI subprocessors, since Cowork's data handling guidance applies here too.

Gotchas

  • A scheduled prompt spends credits every run whether anyone reads the briefing or not. Tidy the Automations view as part of leaver and role-change processes.
  • Two runs of the same question can differ, so a figure that goes into a steering pack should be checked in the admin center first.
  • Owner-outreach prompts are only as good as the owner data. Resources owned by people who have left will surface as ownerless or fail to resolve, which is the finding, but not the email you wanted to send.
  • Showback by department depends on directory data being clean; blank department fields turn into an "Unknown" bucket nobody wants to own.

Consultant notes

  • Position it as a faster way to ask the questions you already ask, not as a replacement for a governance model. The client still needs environment strategy, data policies and a named owner.
  • Run the pilot on a read-only question set first. Once admins trust the answers, the scheduled briefing is an easy win to demonstrate.
  • Put "validate before contacting makers" into the runbook. The first wrongly-accused maker will do more damage to adoption than any report will repair.
  • Keep the credit-cost conversation early; admins who'd never ask for a Power BI licence will happily schedule ten daily prompts.

Worth revisiting when the plugin leaves preview or the credit pricing for Cowork changes.

Was this accurate?