From Prototype to Production: An AI Agent in Your Editor and the Final Build

• A prototype and a production site are not the same: production needs real data, security, tested forms, speed and SEO.
• An AI agent in your editor reads the whole project, changes several files at once and runs checks in the terminal by itself.
• Project rules written in a file (e.g. AGENTS.md) tell the agent which components, colours and structure to use, so the code stays consistent.
• The production layer: database (Supabase), security (RLS), forms (Server Actions), emails (Resend) and payments (Stripe).
• Before launch: a TypeScript check, a build and a manual test on a phone, then push to GitHub and deploy on Vercel.

The agent is fast, but you are responsible for the result:

• Give it a clear, small task and name the files it concerns.
• Ask for a plan before changes, and review the diff before you accept it.
• Have the agent run the checks (tsc, lint, build) and show you the result.
• Write recurring rules into AGENTS.md instead of repeating them in every prompt.

1. Repository with the prototype -> 2. Database and security -> 3. Forms and emails -> 4. Payments (if needed) -> 5. TypeScript and build check -> 6. Push to GitHub and deploy on Vercel -> 7. Connect the domain and Google Search Console.

1. Final check before pushing to production

   npx tsc --noEmit && git push origin main

   Only pushes if TypeScript reports no errors, so Vercel doesn't start a production build from broken code.

Use a prototype to test the idea quickly, and only invest time in the production build once the direction is confirmed. That way you never fully build a site nobody has approved.