Drowning in AI? Start with a CLAUDE.md File
Drowning in AI? Start with a CLAUDE.md File
Everyone asks how to learn AI coding tools "properly." There are hundreds of courses, YouTube videos, and Twitter threads with conflicting advice. The biggest thing that actually helped me was a mindset shift: treat the AI like a junior developer you are managing, not a magic box.
The Junior Dev Mental Model
When you onboard a junior developer, you do not just say "build the settings page" and walk away. You give them context about the codebase, explain the conventions, point out the gotchas, and review their work carefully.
AI coding tools need exactly the same treatment. The difference is that your "onboarding doc" is a CLAUDE.md file in the project root.
What Goes in CLAUDE.md
Start simple. Write down: the project structure, naming conventions, which libraries you use and why, common patterns in your codebase, and things the AI should never do (like using em dashes, or modifying certain files, or changing the database schema without asking).
This file gets read by the AI at the start of every session. It provides the persistent context that prevents the AI from making decisions that drift from your architecture.
Build Incrementally
Every time the AI does something wrong, add a note to CLAUDE.md. "Never use library X, use library Y instead." "Always handle errors with pattern Z." Over time, the file becomes a comprehensive guide that makes the AI increasingly effective.
This is exactly how you would train a junior developer - through accumulated feedback over weeks and months.
Stop Trying to Learn Everything at Once
You do not need to master prompt engineering, fine-tuning, RAG, agent orchestration, and twenty other concepts before you can be productive. Start with one tool - Claude Code, Cursor, whatever - and one CLAUDE.md file. Use it daily. Learn what works and what does not through experience.
The people who are most effective with AI tools are not the ones who took the most courses. They are the ones who built real projects and iterated on their workflow.
Fazm is an open source macOS AI agent. Open source on GitHub.