Haven’t tried Claude Code yet, but I am quite intriqued by it. As I understand it you can define multiple agents and give them different roles, is that correct? How did you set it up?
Some ingredients for successful AI usage for you:
- Claude Code running Opus 4.8 high thinking, not Sonnet. Yeah, you’ll need a decent subscription.
- Or…OpenAI’s Codex running on xhigh thinking. Every bit equal to CC. Less costly.
- Run in a terminal window in Visual Studio Community at the root of your project.
- No JUCE or CMake shenanigans necessary for just using it.
- Use a unit test framework such as GTest. Get AI to integrate it.
- Get the AI to write solid unit tests for work done.
- Create a CLAUDE.md (Claude code) or AGENTS.md (Codex) with instructions on coding style, when to do unit tests, compilation preferences etc.
- Work with the AI to define task requirements first, iterate until you’re happy.
- Get the AI to create a plan from the task.
- Get the AI to spawn to new review agent to challenge and review the plan, fixing any errors if necessary.
- Get the AI to implement the plan, then spawn a new code review agent to challenge and review the implementation, fixing any errors if necessary. Make sure all unit tests pass.
Know what you’re doing. Don’t abdicate responsibility to the AI - you’re accountable. Use it to accelerate.
2 Likes