- Use Cursor or Wind Surf as AI-powered IDEs.
- Prefer Claude 3.7 Thinking or Grok 3 for agentic coding.
- Ensure API keys are correctly set up.
- Define project rules in the IDE to control AI behavior.
- Clarify the problem first -- no code without a clear problem statement. Ask for clarification instead of guessing.
- Start with detailed specs before generating AI code.
- Keep AI requests small and precise to avoid unnecessary changes.
- Run tests frequently to verify AI-generated code.
- Prefer end-to-end tests over unit tests for better real-world validation.
- Monitor AI chat context size, restart sessions when performance drops.
- Prefer simple solutions, avoid over-engineering. Ensure basic correctness before adding abstractions.
- Self-review generated code -- argue against your own solution, check for simpler alternatives. Prefer refactoring over adding code when fixing errors.
- Eliminate code duplication, reuse existing functions where possible.
- Maintain separate environments for DEV, TEST, and PROD.
- Only apply requested changes, avoid modifying unrelated parts.
- Don’t introduce new tech or patterns unless strictly necessary.
- Keep the codebase clean & structured, refactor regularly.
- Never use mock data in DEV or PROD, only for tests.
- Never overwrite .env files without explicit approval.
- Continuously update the README (installation, maintenance, key info).
- Always verify AI-generated code before proceeding.
- Maintain a TODO file (To Do, In Progress, Done) to track AI progress.
- Commit frequently to enable easy rollbacks.
- Use built-in IDE versioning to revert changes when needed.
- Run AI modifications in separate branches before merging to main.
- Never use YOLO mode in production, manually approve critical deployments.
- Stick to popular tech stacks (Python, JS, SQL, etc.) for better AI support.
- Use multiple AI agent windows to develop features in parallel.
- Leverage AI chat history & restore checkpoints when needed.