GitHub Copilot: using AI autocomplete without letting it drive

What it's best at

Boilerplate is Copilot's home turf: the test file scaffold, the data-parsing loop, the config you've written forty times. It also shines at translating a clear comment into a first draft of a function — the clearer the comment, the better the draft.

It suggests what similar code usually looks like, which is exactly why you must review it: usual is not the same as correct for your case.

Working with it, not under it

Write the function signature and a comment describing inputs, outputs and edge cases before letting Copilot suggest the body. You're doing the design; it's doing the typing.

Read every suggestion as if it came from a fast, confident junior colleague. Accept small completions freely; treat multi-line blocks as drafts to interrogate.

The traps

Copilot will cheerfully call functions that don't exist, use deprecated APIs, and reproduce patterns from its training data that your codebase deliberately avoids. It also doesn't know your tests must pass — you do.

Be careful with secrets: never paste credentials into prompts or comments, and check that suggestions haven't interpolated anything sensitive from open tabs.

Plans in short

Copilot offers free and paid tiers with different feature sets, and free access for verified students and maintainers of popular open-source projects. The lineup changes; check GitHub's current plan page for what applies to you.