What happens when GitHub Copilot stops simply helping you write code and starts working directly with Microsoft Power Platform?

In this session I'll take you through a practical journey of setting up an agentic development environment for an existing Dynamics 365 and Power Platform solution using Visual Studio Code, GitHub Copilot, Copilot CLI, Model Context Protocol (MCP), and Microsoft's Power Platform and Dataverse Skills.

You'll learn how Copilot can connect to Dataverse, discover standard and custom tables, understand an existing Model-Driven App, and use those specialist Skills to write and deploy a customised and scalable, bespoke solution for your customer.

But giving an AI agent access to an existing Power Platform environment introduces challenges too. We'll look at what happens when GitHub, Microsoft Entra, PAC CLI and Azure CLI identities don't align, why MCP tools can be discoverable but unavailable to an agent, and why downloading an existing application doesn't necessarily mean every component can safely be round-tripped.

I'll share the practical lessons, guardrails and troubleshooting techniques learned along the way, including why it's so essential to go from read-only discovery to a reviewed change plan before allowing an agent anywhere near modifying your environment.

If you're a Power Platform architect, developer or maker wondering what agentic development actually looks like beyond the demos, this session is for you.

Level: Advanced / 300

Audience: Power Platform developers, Solution Architects, technical consultants and advanced makers.

Core message: This isn't a "look what Copilot generated" session. It's a real-world architecture and engineering session about safely moving from AI assistance to agentic Dynamics 365 and Power Platform development.