I have recently started using Claude Code to handle parts of my day-to-day .NET development — generating boilerplate, reviewing pull requests, and creating task-specific "subagents" for repetitive work like code review or test generation.
So far I've experimented with:
The built-in /agents command to spin up specialized subagents
Manually writing agent config files for more control over behavior
Letting it auto-delegate parts of a refactor to the right subagent
I'm curious how other .NET developers are actually using AI coding agents (Claude Code, GitHub Copilot, or similar) in real projects, not just demos.
Have you set up custom agents for specific .NET tasks — EF Core migrations, unit test generation, code review, etc.?
Do you prefer one general-purpose agent, or several narrow, specialized ones?
Do you let an agent make direct code changes in a larger codebase, or do you keep it strictly to suggestions you review first?
Would love to hear what's genuinely working for people day-to-day.