Let Copilot Agent Mode handle coding, then constrain it
A developer let Copilot Agent Mode do all the coding, found they understood their code less, and moved to using it more narrowly.
Done withGitHub Copilot
- What they did
- The poster initially used GitHub Copilot Agent Mode for nearly everything: tests, refactors, and new modules. After finding that blind delegation made them understand and debug their own code less, they changed to a tighter loop where the agent still proposes work but they interrogate its approach against their architecture and review the output instead of just accepting it.
- How it went
- The reported result was qualitative: they felt the shift from “do everything” to “challenge me” made them a better developer and helped them keep ownership of the codebase. In comments, others said success depended heavily on a well-structured repo, and one claimed full features worked about 90% of the time.
- Worth knowing
- The main catch was loss of understanding: when the agent handled too much, debugging got harder and the developer felt they were slipping into autopilot.
Try it yourself with GitHub Copilot
In my [language/framework] project, use agent mode only for [one narrow task, e.g. writing unit tests for the utils folder]. Don't change files outside [folder or file names]. After each change, explain in plain language what you did and why, so I can follow the code, and stop when [the task is complete, e.g. all tests pass].
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: Reddit · Undated
Related use cases
Run a solo company with eight agent departments
A founder set up CEO, CFO, COO, lawyer, accountant, marketing, CTO, and improver agents that share memory and consult each other.
Iterate on tasks with Copilot CLI autopilot and fleet
The author used autopilot to iterate until done and /fleet to split work across parallel subagents, noting overhead on small tasks.
Assign GitHub issues to Copilot coding agent
A developer assigned a GitHub issue to Copilot and watched it work, but it started from the default branch instead of the intended feature branch.
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Refactor a million-line repo with 1,393 subagents
A Nous Research developer had Hermes dispatch 1,393 subagents over about 19 hours to clean up its own repo, cutting non-test Python source by 34.4% in a merged PR.
Build a mobile app turning photos and voice into reminders
The poster built Loopi, an app that turns photos, text and voice into reminders, notes and expenses, in about 45 days using Replit and AI agents.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.