How to vibe code with AI: 30 tips from Y Combinator

Y Combinator literally dropped the best guide to vibe coding 👇 Here are the 30 tips that will 10x your AI coding results: → Before you start: 1. Don't dive straight into code — work with the LLM to create a comprehensive plan first 2. Put this plan in a markdown file and implement section by section 3. Mark features as "won't do" if too complex, keep an "ideas for later" section → Choose tools wisely: 4. Beginners: Start with Replit or Lovable for visual interfaces 5. Experienced devs: Jump to Windsurf, Cursor, or Claude Code 6. Pro tip: Load both Cursor and Windsurf simultaneously for different iterations → Version control is everything: 7. Use Git religiously — commit after each working section 8. Don't trust built-in revert functions yet 9. If AI gets stuck in loops, git reset --hard and start fresh → Testing strategy: 10. Write high-level integration tests that simulate user clicking through the app 11. Avoid low-level unit tests — focus on end-to-end functionality 12. Implement tests before moving to the next feature 13. Use tests as guardrails for safe refactoring → Debugging: 14. Copy-paste error messages directly into the LLM 15. Ask AI to think through 3-4 possible causes before coding 16. Reset after each failed attempt — don't accumulate "layers of crap" 17. Add extensive logging when debugging complex issues 18. If one model fails repeatedly, switch to a different LLM → Advanced configuration: 19. Write detailed instructions files (some founders write 100+ lines) 20. Download API docs locally instead of pointing to web docs 21. Use screenshots to show UI bugs or design inspiration 22. Try voice input tools like Aqua for 2x faster prompting → Architecture best practices: 23. Keep files small and modular — easier for both humans and LLMs 24. Choose mature frameworks with lots of training data (Rails > Rust/Elixir) 25. Build complex features as standalone projects first, then integrate 26. Use service-based architecture with clear API boundaries 27. Test complex functionality in isolation before adding to main codebase → Non-coding applications: 28. Use AI as DevOps engineer: configure DNS, set up Heroku hosting 29. Generate and resize images, favicons, simple graphics 30. Use AI as teacher: ask for line-by-line code explanations Think of AI as a programming language where you program with words, not code. The best results come from applying professional software engineering practices. -- ⭐ Bonus for non-native English developers: Fluently - AI English coach to improve your English skills. Like a human tutor, but 15x cheaper → https://getfluently.app/