But three months ago that all changed with the release of Claude Sonnet 4 and Gemini 2.5 Pro. These models are something else. If you've tried AI generating code in the past and been disappointed, try again.
Final Takeaways
• Set up strong scaffolding and type-safe context
• Codify conventions in .cursor/rules
• Load all relevant files—including types—into Cursor’s context
• Use planning mode on top-tier models (o3‑pro, Claude Opus 4, etc.)
• Lint, typecheck, and test after each plan execution
• Embrace audio-first prompting
• Enjoy these last few years of human-led coding magic