Let Copilot agent mode iterate on involved coding tasks

A commenter uses Copilot agent mode to pick files, run commands, and iterate until involved coding work is done.

Done withGHGitHub Copilot

What they did

The commenter says Copilot agent mode becomes useful when you hand it involved coding work instead of tiny edits: jobs where it must decide which files to change, run commands, inspect the repo, and iterate until the task is done. In the thread, examples include starting with plan mode to map a technology upgrade and unit-test coverage, then having agent mode implement it, with the user improving results by adding instruction files, prompts, skills, or custom agents.

How it went

The reported outcome is qualitative rather than numeric: several commenters say this works well for multi-file refactors, debugging, compiling, and fixing errors, while small obvious edits still feel slower than doing them manually. No hard performance numbers are given in the thread.

Worth knowing

The main catch is setup and learning time: commenters stress that agent mode improves once you give it repo-wide context and project-specific instructions instead of treating each chat as self-explanatory.

Source: Reddit · Undated