Teams' AI Project Manager is here: How to set it up and why your Copilot license matters

Microsoft has begun rolling out a public preview of an AI-driven Project Manager agent inside Teams that can autonomously create project plans, assign tasks to itself, and generate status reports – all from the transcripts of your meetings and chat conversations. The feature, first detailed by Windows Report, promises to eliminate hours of manual follow-up work, but turning it on requires Microsoft 365 Copilot licenses, a fully enabled Loop environment, and careful governance planning that goes well beyond a simple toggle.

What the agent can actually do

The Project Manager agent isn't a passive suggestion tool. Once activated, it operates inside Planner, integrated directly into the Teams interface, and can perform a range of project-management chores. You can ask it to build a plan from a high-level goal, and it will auto-populate buckets, milestones, and tasks. You can assign it a task, and it will execute that task – producing a draft document, analyzing data, or compiling a status update – and then deposit the results into a Loop page linked to the plan.

Imagine you’ve just finished a 30-minute project standup. The Facilitator agent has captured the transcript and identified that someone needs to update the risk register by Friday. Instead of you or a colleague creating a Planner task and then doing the work manually, you can @mention the Project Manager agent in the task’s comments: “Please draft an updated risk register based on this meeting and last week’s status report.” The agent will queue the task, execute it, and within minutes post a Loop page with a draft document for your review. That’s the promise – meetings that don’t just end, but directly generate progress.

The agent tracks its own progress through states like Queued, In Progress, Needs Input, and Done, and will ping you in Teams when it hits a roadblock or finishes a job. It can also generate one-click deliverables, such as status reports in Loop or Word, drawing directly from meeting notes and plan data. Crucially, the agent never marks a task as complete on its own; a human must review the output and manually close the task. That design choice keeps a human in the loop, a requirement that Microsoft baked in to prevent unchecked automation.

The licensing maze: What you need before you start

If you’re hoping to turn this on for your team, you’ll need more than just the latest Teams update. The Project Manager agent sits at the intersection of several Microsoft 365 products, and the licensing prerequisites are layered.

First, every user who wants to interact with the agent – assigning it tasks or asking it to execute work – must have a Microsoft 365 Copilot license. That’s the non-negotiable entry ticket. Second, the agent’s output is stored as Loop pages, which are housed in SharePoint folders tied to the plan. The agent’s output lands in Loop pages, which are essentially collaborative canvases stored in SharePoint. This design means every agent-created artifact is automatically subject to your existing SharePoint permissions and compliance policies, but it also means that if your tenant has Loop turned off or restricted to certain users, the Project Manager agent won’t be able to produce its deliverables. Additionally, the agent uses Loop to request clarification from humans – if it doesn’t understand a prompt, it will post a Loop component asking for more details, and then wait for a response before proceeding.

Third, while basic Planner functionality is included with most M365 plans, premium features like task dependencies, baselines, and advanced timelines require a Planner premium license (Planner Plan 1, Planner and Project Plan 3, or Plan 5). Users without that premium license can still collaborate on plans but won’t get the richer scheduling tools.

In short: a Copilot license is mandatory for the agent itself, Loop is effectively mandatory for full functionality, and Planner premium is optional but unlocks enterprise-grade project management. And because the agent is in public preview, your tenant must also be opted into the Teams public preview channel – either tenant-wide or by giving users the ability to switch it on individually.

Setting up: A 5-step checklist for IT

For administrators, the rollout path is fairly linear, but skipping a step will lead to a stalled deployment. Here’s what you need to do:

1. Confirm Microsoft 365 Copilot license assignment. Audit your user base and assign Copilot licenses to anyone who will need to interact with the Project Manager agent.
2. Enable Microsoft Loop. In the Microsoft 365 admin center, ensure Loop is turned on for the tenant. Verify that users who create plans can also create Loop workspaces, since the agent will spin up a workspace and SharePoint folder for each plan.
3. Decide on Planner premium licenses. If your teams rely on dependencies, baselines, or advanced Timeline views, procure and assign the appropriate Planner premium licenses.
4. Turn on Teams public preview. Either set the tenant-wide preview policy to allow the feature or instruct users to opt into public preview from their Teams client settings.
5. Train your users. Because the agent writes its output to Loop pages, users need to know how to find those pages, provide feedback when the agent asks for input, and regenerate deliverables. Without training, the agent’s “Needs Input” state will become a bottleneck.

Governance guardrails: Security, compliance, and agent identities

The Project Manager agent is part of a broader push by Microsoft toward an agent-first collaboration model, which introduces new governance considerations. Microsoft has been building a set of identity and compliance controls for this world of persistent AI teammates.

Each agent can now have an identity in Microsoft Entra, called Entra Agent ID, which allows administrators to manage it like a user or service principal – applying access policies, auditing activity, and even removing it when necessary. This reduces the risk of rogue agents proliferating unchecked. Agent output, including Loop pages and generated documents, inherits the data protection settings of the underlying SharePoint and Loop environments, and Microsoft Purview can be configured to extend sensitivity labels and data loss prevention (DLP) policies to those artifacts. Microsoft Purview can apply sensitivity labels to those Loop pages and generated files, and DLP rules can scan the content for sensitive information. But because the agent is a new type of actor, IT must ensure that Purview policies are configured to recognize agent-created content and apply the same restrictions they would to any other file in SharePoint. This might require updating existing DLP rules to include the specific document libraries or sites where agents operate.

The company also touts its Model Context Protocol (MCP) for secure agent-to-agent communication, ensuring that when agents call each other or external services, they respect tenant-level security controls.

However, these governance primitives are still new. In a preview scenario, IT teams should validate that DLP rules and retention policies actually apply to Loop pages created by the agent. They should also treat any fine-tuned models – if they plan to use Copilot Tuning – as separate, governed artifacts that may contain sensitive data, and include them in compliance reviews. Microsoft states that customer data is not used to train its public models, but the tuning process itself can produce models that require careful access control.

Where things could go wrong: Limitations to watch for

During early testing and community feedback, a few pain points have emerged that administrators should prepare for.

The agent’s reliance on Loop and SharePoint means its uptime is tied to those services. If your organization has strict retention policies that delete Loop pages after a certain period, the agent may lose context and be unable to re-run tasks that reference deleted pages. Syncing between meeting follow-ups and Planner can also be flaky, especially with recurring meetings or channels that have complex policy settings. Microsoft’s own Q&A forums note that some action items don’t reliably appear in Planner, and troubleshooting often involves digging into meeting metadata.

Language support adds another layer of uncertainty. While meeting-focused agents like Facilitator and Interpreter support multiple languages (Interpreter covers nine languages at preview), the Project Manager agent’s input and output interfaces may initially default to English during staged rollouts. If your teams work primarily in another language, test the agent’s performance on non-English plans before committing.

Finally, cost segmentation is a real concern. Not everyone on a project team will necessarily have a Copilot license, meaning some members can assign work to the agent while others cannot. This creates uneven automation within a single plan and can confuse workflows if not managed explicitly.

Bottom line: When to deploy

The Project Manager agent represents a meaningful leap for Teams-based collaboration. It can transform the drudgery of meeting follow-ups and status reporting into an automated pipeline, freeing project managers to focus on higher-level decisions. But the payoff is conditional on a solid foundation of licensing, infrastructure, and governance.

For organizations that already have Microsoft 365 Copilot rolled out and Loop enabled, flipping the switch on a pilot program is a logical next step. Start small: pick a cross-functional team, provision the necessary licenses, run the agent on a handful of projects for four to six weeks, and measure whether the time saved on report generation and task execution justifies the licensing cost. For heavily regulated industries or those still evaluating Copilot, this is the moment to begin mapping out your agent governance strategy – because whether or not you enable Project Manager agent today, agents are coming to the entire Microsoft 365 suite, and you’ll need a playbook.

Outlook: What’s next for Teams agents

Microsoft’s multi-agent vision doesn’t stop here. The company is rapidly expanding its portfolio of specialized agents inside Teams – Facilitator for meetings, Knowledge Agent for SharePoint, Channel and Community agents for persistent chat – and the underlying platform components (Entra Agent ID, Purview, MCP) are being built out in parallel. Expect these agents to become more tightly integrated, with the Project Manager agent learning from past projects and potentially orchestrating other agents to complete subtasks. The public preview may be rough around the edges, but it signals a shift toward AI that doesn’t just answer questions – it actually does the work. For IT teams, the message is clear: start experimenting now, but build your governance framework first.