VS Code Agent-First Development Series
VS Code Agent-First Development Series
A 6-episode official curriculum from the Visual Studio Code team on working with AI coding agents (GitHub Copilot). Published April 2026. Total runtime ~44 minutes.
The Core Idea
Agents aren't magic — they follow a predictable formula: harness (VS Code + Copilot), model (choice + thinking effort), context (files, MCP, # references), tools (terminal, search, delegation), and prompts (detailed but not over-prescriptive).
Episode Map
| Ep | Focus |
|---|---|
| 1 | The 5-part formula explained — harness, model, context, tools, prompts |
| 2 | First agent session — approval levels, tool calls, context management |
| 3 | Reviewing and controlling changes — steer, fork, checkpoints, diffs |
| 4 | Where agents run — local, cloud, CLI; Ask / Plan / Agent modes |
| 5 | Debugging agents — per-session logs, raw LLM I/O, /troubleshoot |
| 6 | Full build demo — Plan mode → Autopilot, steering mid-build |
The series builds cumulatively: each episode layers one or two capabilities on top of the last, ending with an end-to-end app build.
See Also
- AI Agents & Automation MOC — full agent tool landscape
- Safe AI Workflows - Mitigating Hallucinations and Overreach — safety guardrails for agent work