Copilot may already finish lines and draft whole functions for you. Agentic coding starts when you give it a bounded task, let it work across files, and inspect the result as carefully as you would a teammate's pull request.
That changes what you need to be good at. An agent that can read files, run commands, and open pull requests needs a clear boundary. Its result needs independent evidence. Throughout this module, you'll keep the diff, test output, or traced source close enough to check before you trust the summary.
You will use agent mode in the editor and assign issues to the cloud coding agent. You'll also move recurring rules into instructions, prompt files, custom agents, and skills. Later lessons connect external tools through MCP, organize source material in Spaces, and use Spark to prototype an application. The final lesson takes the same review discipline into team workflows and organization controls.
These interfaces change quickly. The working method is more durable: define the task, control what the agent may do, and verify what it returns. Bring a disposable repository so you can try the patterns without putting real work at risk.
What you'll walk away with
Direct local agent mode and the cloud coding agent within reviewable boundaries
Customize Copilot with instructions, prompt files, custom agents, and skills
Connect and govern MCP servers, then use Spaces and Spark for context and prototypes
Scale code review and agentic workflows on evidence-first standards
Govern Copilot across an organization and choose models deliberately
