For the work that is mostly typing

What it is

GitHub Copilot is an AI assistant built into the code editor. Its best-known job is completion: as you type, it suggests the rest of the line, or the next few lines, based on the file you are in and what you appear to be doing. Accept it with a keystroke, ignore it and keep typing. It now does more than that, but completion in the editor is where it earns its place here.

Where it fits

A surprising amount of engineering is not hard thinking. It is typing something you already know how to write.

• Boilerplate. The fifth similar function, the matching test case, the obvious next field in a data structure.
• Repetitive edits. Making the same kind of change down a long file, where once the pattern is clear the rest is keystrokes.
• Familiar syntax in an unfamiliar corner. Remembering the exact form of a rarely used command or library call without leaving the editor to look it up.
• Staying in flow. Small suggestions that save a trip to the documentation and keep attention on the work.

Where it does not

Being fair about this is the useful part.

1. It sees the file, not the estate. Line completion is good at the next few lines and weak at questions like "what else breaks if I change this?". Work that spans many files, servers or systems is better done by a whole-task assistant such as Claude Code.
2. Plausible is not correct. A completion that looks right is easy to accept without reading. On anything that matters — security, data handling, configuration — every suggestion gets read as if a stranger wrote it.
3. It does not know your conventions unless they are visible. In a project with strong local patterns, suggestions can drift towards how code is usually written rather than how yours is written.
4. It does not run anything. Completion has not tested what it suggested. The tests and the review still have to happen.

Why I keep it alongside Claude Code

The two tools answer different questions. Claude Code takes a whole task — investigate this, change that across the project, verify the result — and works through it. Copilot helps with the line you are typing right now. Using a whole-task agent for a one-line edit is slow; using line completion for a system-wide change is the wrong shape entirely.

Keeping both is also part of not betting the estate on one vendor. Each tool earns its place by being the better fit for some kind of work, and that judgement gets revisited as the tools change.

What I would do for a client

For a development team, in-editor completion is often the easiest AI tool to adopt, because it changes very little about how people already work. The questions worth settling first are about governance rather than features: which repositories it may be used in, what your licence terms say about code and data, how suggestions are reviewed, and whether the productivity gain is real for your team rather than assumed. I would help set that up and measure it, not just switch it on.

Questions people ask