If you’ve been working inside Claude Desktop for the last year — attaching project files, asking the model to think through architecture decisions, treating the conversation window as your working surface — the shift to VS Code with GitHub Copilot is going to feel strange at first. The model is largely the same. The room around it is completely different. Copilot lives inside your editor, knows what file you’re looking at, and integrates with your code in ways that don’t map onto a chat-style interface.
This note is an attempt to make that transition fast. It’s written for engineers who’ve already done the conceptual work of integrating an AI assistant into their workflow and are now changing which one. The conceptual cost should be small. The setup cost — if you don’t know what to expect — can be surprisingly large.
What it covers
Eight short sections, about twenty minutes to read.
§1 — What is actually changing. The architectural difference between Claude Desktop and Copilot. What you keep, what you give up, what becomes possible.
§2 — A useful analogy. A two-paragraph mental model that makes the rest of the document click.
§3 — The three modes. Inline completion vs chat vs agent mode. When each one is the right tool.
§4 — Project knowledge, in a file. The Copilot equivalent of Claude Desktop’s project files — but expressed differently and with different gotchas.
§5 — What happens when you press Send. The lifecycle of a Copilot request. What gets sent, what doesn’t, what affects the response.
§6 — Files, images, and memory. The handling of context that isn’t pure text — how to feed Copilot a screenshot, a PDF, a directory.
§7 — A small example, end to end. Walking through a representative task with both tools, side by side.
§8 — What to do on day one. A concrete checklist for the first day of the transition.
Read it
← Back to Autonomy