Overview
Copilot stopped being a suggestion box a while ago. This session looked at the GitHub Copilot agent the way you would look at a new team member — what it can be handed, how it plans and edits across a repository, and where a human still has to stand in front of the merge button.
What we covered
- Agent mode in VS Code — how a single prompt turns into a multi-file plan, edits, terminal commands, and a self-correcting loop.
- The coding agent on GitHub — assigning an issue to Copilot, watching it work in its own environment, and getting a draft pull request back.
- Steering it —
copilot-instructions.md, per-path instruction files, and the context that decides whether the output matches your codebase or fights it. - Tools and MCP — connecting the agent to real systems so it can read issues, query docs, and act beyond the editor.
- Reviewing agent output — reading a diff you didn’t write, Copilot code review, and the checks that keep the branch honest.
- Where it breaks — long refactors, ambiguous requirements, and the tasks still cheaper to do yourself.
Who attended
Developers and leads adopting agentic AI on real repositories. 110+ attended.