Deploy, Convert & Scale

From Vibe-Coded Prototype to Production: The Complete Checklist

A vibe-coded prototype is ready for production when you own the code, secrets are server-side, data access is locked down, critical flows are tested, deployments are repeatable, failures are visible and someone is accountable for operations. This checklist works whether the prototype came from Lovable, Bolt.new, v0, Replit, Base44, Cursor, Claude Code or any other AI tool.

neooutput editorial team 4 min read Auf Deutsch lesen

1. Ownership and source control

First make sure the code lives in a Git repository you control, with history, access management and a protected main branch.

Many AI builders sync to GitHub or allow export; if yours does not on your current plan, decide now whether that is acceptable for the business. Also own the domain, the database account, the email sending account and the payment account in the company's name, not a freelancer's or an individual employee's.

  • Repository in a company GitHub or GitLab organization.
  • Domain, DNS and cloud accounts owned by the company.
  • At least two admins with MFA on every critical account.
  • A README explaining how to run, build and deploy.

2. Security essentials

Before real users arrive, remove secrets from the frontend, enforce authorization on the server and enable database access rules.

Search the code and Git history for API keys, rotate anything exposed, move third-party calls into server functions, and check that every table or collection has access rules (for example Supabase RLS or Firestore Security Rules). Test with two ordinary accounts whether one can see or change the other's data. Add rate limits to login, sign-up and any endpoint that costs money.

3. Data, privacy and compliance

Know what personal data you store, where it lives, who can access it and how you would restore it after a mistake.

Write a short data inventory, enable automated backups and test a restore once. Separate development, staging and production data so AI agents and tests cannot touch live records. For EU users, update the privacy policy, sign data processing agreements with providers, configure cookie consent where required and plan how you handle deletion and access requests.

4. Testing and quality gates

Protect the flows that make or lose money with automated tests, and run them on every change.

You do not need full coverage on day one. Start with end-to-end tests for sign-up, login, the core action and payment, plus unit tests for pricing or permission logic. Add type checking, linting and a build step to CI, and make them required before merge. AI agents become far more reliable when they can run these checks themselves.

  1. Add CI with install, typecheck, lint and build.
  2. Write end-to-end tests for the three to five critical journeys.
  3. Add tests for authorization rules.
  4. Make CI required on the main branch.
  5. Run the suite before every production deploy.

5. Deployment and environments

Production deploys should be automatic, repeatable and reversible, with a staging environment that mirrors production.

Connect the repository to your host, deploy main to production, give pull requests preview deployments and keep environment variables per environment. Manage database schema changes as migrations in Git. Practice a rollback once so it is not a first attempt during an outage.

6. Monitoring and operations

Set up error tracking, uptime checks, logs and alerts, and decide who responds and how.

An error tracker with source maps, an uptime monitor, searchable logs and cost alerts at every provider cover most early risks. Write a one-page runbook: where to look, how to roll back, whom to contact at each provider.

Also add cost monitoring: budget alerts at the hosting provider, the database, the email service and every AI API. A bot attack or a loop in generated code can create a large bill overnight, and an alert is the cheapest way to catch it.

7. Performance, SEO and team process

Finish with performance basics, search visibility for public pages and a working process for ongoing changes.

Check Core Web Vitals on a mid-range phone, paginate lists and index database queries. If organic traffic matters, serve public pages with server rendering or prerendering. Finally, agree how the team changes the app: who prompts which tool, how changes are reviewed, and which rules files (CLAUDE.md, AGENTS.md, Cursor or Windsurf rules) document the conventions for AI agents.

If a mobile app is on the roadmap, decide early between a PWA, a WebView wrapper such as WebViewGold, Capacitor or a native rewrite with React Native or Expo, because the choice affects authentication, push notifications and how you structure the backend API.

Frequently asked questions

What is the biggest risk when launching a vibe-coded app?

Broken access control and exposed secrets. AI-generated apps often work perfectly in demos while letting any logged-in user read other users' data or exposing API keys in the browser.

Do I need to rewrite a vibe-coded app before launch?

Rarely. Most prototypes need hardening, tests, proper deployment and monitoring rather than a rewrite. Rewrite only parts that measurably fail requirements.

How long does it take to make a prototype production-ready?

It depends on the app's complexity, data sensitivity and existing quality. A focused audit quickly shows which checklist items are already in place and which need work.

All insights