Experimenting with AI subagents

This article details the author's experiment using AI subagents (via Copilot CLI) to analyze a codebase, triage GitHub issues, and automate fixes. The author describes prompting Copilot to identify improvements and bugs, then creating GitHub issues with labels. After triaging, four solid issues remained, and the author used sub-agents to autonomously fetch issues, create branches with git worktrees, implement fixes, write tests, commit, and push PRs. The article covers practical tips like using gh for write permissions and avoiding agent conflicts with git worktrees. It is a technical tutorial focused on AI-assisted software development workflows.