How I Use AI

Just talk to the AI and let it build. No specs, no plans, just vibes.

Plan first with Claude, then switch to auto-accept for one-shot execution.

Autonomous iteration loop — set success criteria and let Claude run until done.

Write a detailed spec first, then let the agent implement it faithfully.

Write failing tests first, then let the agent make them pass. Red → Green → Refactor.

Invest heavily in CLAUDE.md files to make every session smarter.

Run multiple Claude sessions simultaneously on different tasks.

Use extended thinking for complex architectural decisions and tricky bugs.

Break complex tasks into isolated subtasks using the Task tool for context isolation.

Point Claude at a GitHub issue and let it produce a complete PR.