Trusted IT Group Blog

Orchestrating human-AI collaboration in Microsoft Planner

Written by Yuri Frid | Nov 21, 2024 5:59:30 PM

How Microsoft Planner is shaping the future of work management

 Project Manager agent is an end-to-end work orchestration experience in Planner that leverages generative AI to streamline the journey from idea, to plan, to done. This is the first time the groundbreaking automation capabilities of Microsoft AutoGen will be available to customers at scale.

The new Project Manager agent will be rolling out to public preview in the Planner app in Teams in the coming weeks. To explore these capabilities, customers are required to have a Microsoft 365 Copilot license and also need to ensure their current Microsoft 365 licensing allows them access to Microsoft Loop.

Introducing the Project Manager agent

 

A team of agents working together

Project Manager agent in Planner orchestrates multiple agents like a project manager. Working directly with task-specific execution agents, Project Manager agent assures tasks are completed efficiently. This GenAI-powered system uses task decomposition, planning, and a multi-agent orchestration to enhance team productivity.

The system brings together specialized agents, each with their own expertise, to work collaboratively on various tasks with humans in the loop to review and guide outcomes. These agents communicate and cross-check each other's work, much like a team of experts refining each other's contributions.

 

At Microsoft Research, pioneering work on AutoGen has demonstrated the benefits of multi-agent systems. Project Manager agent puts these ideas into practice, streamlining completion of complex tasks:

  • Specialized Expertise: Each agent is a specialist, ensuring high-quality outputs tailored to specific tasks. Specialized agents make output more reliable and predictable while preserving the dynamic nature of agentic systems.
  • Team Collaboration: Agents communicate and cross-check each other's work, much like a team of experts refining each other's contributions. Adversarial agents are specifically designed to challenge each other's work to produce more refined and detailed results.
  • Problem Solving: AutoGen research demonstrates that multi-agent systems significantly enhance problem-solving capabilities, enabling them to tackle more complex tasks effectively. For instance, the integration of multiple agents improves performance in math problem-solving (MATH), real-world reasoning (AFLWorld), and sensitive coding tasks (OptiGuide).
  • Contextual Awareness: Certain agents provide context from your team's existing content, integrating historical data and insights into the current plan. Research on AutoGen shows that multi-agent systems facilitate improved contextual awareness and understanding by integrating multiple retrieval mechanisms, enabling a more interactive and dynamic approach to information retrieval.

While in preview, Project Manager agent has access to a small set of "built-in agents" that are tuned to complete common information work tasks. Project Manager agent is scalable, and we are excited to expand the set of available agents and capabilities over time.

Human-AI collaboration

Whether asking Copilot a simple question or automating a complex workflow, teams in every industry are accelerating work with Generative AI. Collaboration is already a pain-point for many teams and the introduction of AI into workflows can further confound the problem. We believe that Planner’s familiar and accessible collaborative work management concepts and tools can effectively streamline human AI collaboration. Project Manager agent leverages Planner to plan, execute, and iterate agentically in an environment where the human team maintains control.

Work Breakdown – When presented with a complex goal or task, Project Manager agent identifies a set of steps to achieve it. This structured approach not only breaks tasks into manageable pieces but also distinguishes work suitable for autonomous agents from higher-value or complex tasks requiring human input. Under the hood, Project Manager agent will be collaborating with multiple agents to deliver the desired outcomes. Complex task execution requires creating a plan, intelligently sequencing actions and dependencies, and maintaining a memory system to store the status of individual steps and their outcomes.

Assign to Agent – The user is always in control of exactly what and how much work is assigned to Agents. Project Manager agent is flexible and can handle single tasks or orchestrate longer flows. At any time, team members can pause agent work and make manual adjustments as needed.

Human in The Loop – During task execution, Project Manager agent can identify situations where more information or context is required from the team. In such situations, Project Manager agent will compile a list of questions that, when answered, will unblock the Agents working on the task. If the system lacks the tools to complete a task altogether, Project Manager agent can provide advice and assistance to the team member working on the task.

Integrating Feedback – Project Manager agent won’t always complete tasks perfectly on the first attempt. Once initial agent work is complete, the team can review the task and leave comments on the content. When the review is complete, Project Manager agent will orchestrate iteration on the content, incorporating the feedback.

Safety and privacy

Generative AI systems have unprecedented capabilities, which introduces novel risks into Microsoft's products and services. The Planner team is committed to proactive and continuous mitigation of these risks. We employ a handful of key strategies to ensure that we responsibly deploy generative AI:

  • Impact Assessment: We audit new generative AI products in accordance with Microsoft’s responsible AI standard, identifying harms the system could potentially cause.
  • Redteaming: We thoroughly evaluate new products for risks identified in the impact assessment throughout the development process. Ahead of release, our redteaming results are reviewed by a joint Microsoft-OpenAI Deployment Safety Board, as outlined in Micrisoft’s AI Safety Policies.
  • Harm Mitigation: Project Manager agent has multiple redundant checks to prevent harmful content, including harm-specific classifiers and a dedicated responsible AI agent.

Additionally, your data remains private when using Project Manager agent. Customer data is governed by the commitments we make in the Microsoft’s Data Protection Addendum, Microsoft’s Product Terms, and the Microsoft Privacy Statement.  

The future of work management

As AI technology continues to evolve, we can expect even more advanced features and capabilities to be integrated into Planner. The vision is to create a truly intelligent work management platform that not only supports project managers but also empowers teams to achieve their best work.

The orchestration of human-AI collaboration with AutoGen in Planner represents a significant leap forward in the field of work management. Project Manager agent is a testament to the potential of AI to transform the way we work, making it more efficient, productive, and collaborative. As we look to the future, we are excited about the possibilities that AI holds for work management and beyond.

Book a Free Consultation to Explore How New PM Agents Integrate Seamlessly with Your Workflow!