Building AI Agents with Copilot Studio & Dynamics 365

Disclaimer: This course contains the use of artificial intelligence.

In expert-level agent architecture, two answers usually stay defensible until one constraint decides between them. This course teaches you to find that constraint, quickly and reliably, across two areas where it matters most: designing agents and keeping them healthy once they are live.

Every lesson is built around a single architecture decision and the fact that settles it. Why the trigger, not the prompt, picks the architecture. Why a Copilot feature that is enabled in Dynamics 365 is not yet deployed. Why a topic can sit in your agent and never fire. Why an impressive resolution rate may be counting silence as success. You learn the platform behaviour these decisions turn on, and you learn to reason with it the way an architect does.

What’s inside

• 6 modules: 24 focused lessons plus a one-pass summary at the end of each module (about 2 hours of video)
• Expert-style scenario quizzes for agent design, Copilot across Dynamics 365, prebuilt AI features, and monitoring. Every question explains why each wrong option fails and names the deciding constraint
• Modules are numbered 5 to 10 to match their place in the wider learning path

How the course is organised

• Designing agents: how the trigger picks the architecture, why it is also the attack surface, and which steps must be repeatable enough to leave the prompt for a deterministic flow
• Copilot across Dynamics 365: teaching Copilot your business vocabulary, the gap between enabled and deployed, and scoping one agent per workstream
• Copilot Studio design and the Well-Architected view: topics that never fire, descriptions as routing logic, which tools belong in the agent, and who owns custom code
• Extensibility: subscribing instead of building, choosing sign-in by channel, controlling the tools an MCP server exposes, and when Computer Use should take the keyboard
• Orchestrating prebuilt AI features: grounding answers in the right product, permissions spread across admin centres, and the security role named for the job
• Monitoring, telemetry, and tuning: live health checks, metrics that flatter you, choosing the right denominator, reading paired signals, treating silence correctly, and setting up alerts

The scenarios follow one fictional company, Kestrel, through the decisions a real programme faces, so each fact arrives with the situation where it matters.

Microsoft’s platforms change quickly. Every fact in the course is sourced to Microsoft documentation and dated, so always check the current documentation before you build.