Microsoft is all in on the Generative AI trend. And to be more precise: the Agentic AI trend.

Agentic AI refers to a class of artificial-intelligence systems that go beyond just responding to human commands or analysing data. These systems set goals, plan sequences of actions, use tools or other systems, and act autonomously (i.e., with minimal human supervision) to achieve outcomes.

Microsoft has built six key agents, one of which is the core topic of this article:

• Facilitator agent, handles meeting-moderation, real-time notes in Microsoft Teams.
• Project Manager agent, supports project tasks, plan creation, execution in Microsoft Planner.
• Interpreter agent, real-time speech-to-speech interpretation in meetings.
• Analyst agent, transforms data into insights within Copilot Chat.
• Researcher agent, synthesizes information, content generation.
• Channel agent, agents scoped to Teams channels, summarizing tasks and status.

Additionally: Microsoft mentions that the SharePoint agent will be created for every SharePoint site.

The Project Manager Agent

This agent was introduced in an early version during last years Ignite. The agent is accessible to anyone with a M365 and a M365 Copilot license. This combination provides you with the following option in the “new plan menu”:

In the beginning, this agent wasn’t much to write about. But times change quickly in the world of AI. And it’s about time we get an update on two features the PMA can do which are going to be gamechangers.

1.   Agent to Agent Communication

One of the flagship agents at Microsoft is the Facilitator agent. It’s proximity to the users during meetings makes it the ideal agent to offload some of the tedious tasks such as note taking, schedule adherence, follow up suggestions and… task creation!

This is where the Facilitator agent hands over a “task” to the Project Manager Agent. Which will create a task or even set of tasks within the team that is created specifically for this meeting.

Creating a new Team for each meeting might be undesirable in the future. But this is the current setting for these agents to operate in. It’s understandable as well from a governance perspective. You want all users in the meeting to have access to the tasks/meeting notes/recording. But you also want to limit access for anyone outside that group of people. Added complexity: there might be external users in that group as well.

I’m expecting a lot of users will start interacting with the facilitator agent soon. And while they are, they might even have interacted with PMA or even Analyst Agent without even knowing they did.

This brings a future perspective within our grasp. We can now govern a team of agents to do tedious tasks and assist brainstorming complex ones. And it is all done in the tools we use most (Teams, Outlook etc).

2.   A Different Interface

Creating a new Plan with Project Manager has changed. Now, we get an extra option to allow the PMA to “access files from this group”. The goal of this is obviously to ground the agent with information from the group. It’s also a requirement to have the plan linked to a group when using PMA. The reason behind that is the close proximity to other agents and the data grounding needs to happen somewhere.

Make sure to uncheck the option if you are adding this schedule to a company wide very diverse team, because it will likely not be able to make heads of tails from the data inside.

Once you are inside Planner, the PMA shows up differently as well. No longer a separate tab (like I showed in my introduction video), the Agent is now an integrated part of your scheduling needs.

This little icon on the bottom right of the screen is also called a Dynamic Action Button (DAB) or Flexible/Fluent Action Button (FAB).

It’s the main point of contact when interacting with the PMA on a schedule. When you click on it this menu shows:

This is a familiar interface that we know from other Copilot like chats. It is also an integrated part of the goals creation we knew from the original PMA. If I click on the first suggested prompt for instance I get the following result:

Follow up prompts 1 and 3 seem interesting, and once I asked the agent to provide the values my schedule, I get an additional option to “Add these actions as tasks in my plan.”

The PMA still struggles with more advanced actions such as durations, and effort. And when prompting

Look at the current actions, create useful buckets for these actions and place all tasks in their relevant buckets.

It didn’t come up with a useful response either. Rather it got stuck in the “applying the finishing touches”, we have seen this before on many such prompts.

Assigning tasks to the PMA will always keep the “Human In The Loop”. The PMA won’t complete a task it’s assigned to, but it will send you a message once it’s done with the (preparation) work. When the agent completes its task there’s a lot of text generated in a Loop page that the agent & human team members can expand on.

Any Loop page can be added to a Teams channel as well:

It might be a good idea to verify the agent on hallucinations though. As to my knowledge, Context& doesn’t have many TikTok followers 😉.

Final notes

That’s it, two recent changes to the PMA that you might want to explore yourself. I’m curious to know, have you worked with the agent before? I would love to get some real-world examples of people using this agent for their scheduling needs. Feel free to comment on the MPUG article or reach out to me on the MPUG forum instead.