🤖

Building Co-Pilot Agents in Microsoft 365

Sep 4, 2025

Summary

  • The session provided a comprehensive, step-by-step walkthrough on building co-pilot agents for various business scenarios using Microsoft 365 tools.
  • Key topics included agent creation methods, licensing and pricing models, configuration best practices, grounding agents with knowledge, and sharing agents with teams.
  • The tutorial emphasized practical tips, highlighted licensing impacts on billing, and included recommendations for getting started efficiently and safely.
  • Resources and references to additional tutorials and downloadable materials were mentioned for further support.

Action Items

  • No specific due-date – Amy: Upload and share the downloadable PDF template and video link for agent setup and licensing comparison.
  • No specific due-date – Amy: Provide a follow-up tutorial on advanced actions for co-pilot agents in the co-pilot Studio web app.

Agent Creation Overview

  • Three main methods to create agents were outlined: within SharePoint document libraries, in the co-pilot app as declarative agents, and using the co-pilot Studio app for advanced customizations.
  • It’s recommended to begin with the simplest approach (declarative agents in Microsoft 365 chat) and iterate as familiarity grows.
  • Agent creation permissions are organization-controlled, and a co-pilot Studio license (included in Microsoft 365 Standalone or as a paid or pay-as-you-go model) is required.

Licensing and Billing

  • Users can try agent creation for free; full features and billing depend on license type.
  • Metered billing applies when agents are grounded on work data or use autonomous actions, unless users have a Standalone co-pilot license.
  • $200 per month covers up to 25,000 messages, or pay-as-you-go at $0.01 per message; referenced another tutorial for deeper licensing details.

Agent Configuration & Best Practices

  • Customization includes agent icon, background color, name, and a clear, functional description.
  • Instructions are critical: must define purpose, guidelines for communication, restrictions, handling errors, feedback, non-standard terms, and closing remarks.
  • Good instructions produce good outcomes ("garbage in, garbage out"), and using chat history can help refine configuration.
  • Knowledge grounding can use up to 20 sources, such as documents and SharePoint sites; agents inherit user's permissions to data.
  • Caution is urged with enabling web content grounding; better to use curated documents for critical topics for accuracy and billing clarity.
  • Advanced capabilities like code interpreter (Python) and image generation can be toggled on as needed.

Starter Prompts and Testing

  • Up to six starter prompts help users interact with the agent efficiently, ideally tied directly to the defined skills in instructions.
  • For each skill, provide clear task steps, goals, and concrete examples of expected output.
  • Regular and thorough testing is encouraged using the preview pane; users should stretch the agent’s responses to ensure reliability.

Deployment, Sharing & Collaboration

  • Agents can be pinned for easy access and shared via Teams channels/posts or added to team homepages with quick links.
  • For editing, it’s recommended to use chat history rather than direct edits in the describe tab, to maintain a record of changes and logic.
  • Collaboration features allow tagging agents in chat to use them in parallel with other conversations.

Decisions

  • Start agent creation with declarative agents in Microsoft 365 chat — Rationale: It is the quickest, most accessible method and allows for incremental improvement.
  • Ground agents on curated internal data for critical use-cases rather than broad web content — Rationale: Ensures accuracy, appropriate billing, and control over information sources.

Open Questions / Follow-Ups

  • Will there be organization-wide guidelines on when to enable advanced capabilities (e.g., autonomous actions, code interpreter) for agents?
  • Confirmation pending on the timeline for fuller rollout of autonomous actions in the co-pilot Studio web app.