AI Learnings

Base starting point: agenticoding.ai

Addendum

Do not blindly trust AI output. The AI may achieve the goal but introduce dead code or break other areas. It can be 90% right in a single iteration, but after multiple iterations the correctness can degrade and technical debt can grow. Clean code and owning the architecture is essential.

Guidance

• Use AI output as if you wrote it yourself. Verify, refactor, and test.
• Ask critical questions and troubleshoot together with the AI.
• Avoid delegating implementation you cannot reason about or validate.
• If the session drifts, restart from a clean commit rather than forcing fixes.
• Reduce error rate with strict typing (TypeScript strict, Python pydantic, etc.).
• Add tests for bugs to prevent regressions.
• Start and end work from green tests.
• AIs are great at tests, but validate that tests reflect correct system behavior.