AI-Built App Production Checklist: What to Check Before Real Users and Payments
A practical launch checklist for apps built with Lovable, Bolt.new, Replit, v0, Base44, Emergent, Google AI Studio or Cursor: access, data, integrations, recovery and operations.
On this page
An impressive demo answers one question: can the intended workflow work in favourable conditions? A launch review asks what happens with the wrong user, repeated requests, an unavailable provider, lost data and a failed release. The same standards apply whether the code was written by hand or with an AI app builder.
This checklist is the common core. For the details of a specific tool, start with its guide: Lovable, Bolt.new, Replit, v0, Base44, Emergent, Google AI Studio or Cursor.
1. Access and data
- Test with two accounts and two organisations. Try to read, edit, download and export the other account’s records, including by guessing resource URLs. Authorisation must be enforced on the server, not by hiding a button. See the OWASP authorization guidance.
- Check the account lifecycle. Password recovery, session expiry, administrator access and removing a team member who has left.
- Look for secrets. Search the repository, built assets, logs and public responses for keys and tokens. If a credential was exposed, replace it; deleting the visible string is not enough.
- Know your personal data. Write down what you store, who can access it and how it is deleted. Get compliance advice for your actual product and markets.
2. Integrations and failure
- Repeat callbacks and submissions in the provider’s sandbox. One intended action must not create two charges, two orders or two messages.
- Make dependencies fail on purpose. Slow down or switch off a provider and check the user sees a usable error, and that enough is recorded to reconcile later.
- Limit abuse. Public forms, uploads and expensive AI calls need server-side rate limits and size limits.
- For AI features, record the provider, model, configuration and date of your evaluation, and define when the assistant must decline or hand over to a person.
3. Deployment and recovery
- A second engineer can build and run it from source and documented steps, without the original builder’s laptop or account.
- Staging is separate and holds no production customer data or production write credentials.
- Restore a backup into an isolated environment and check real records. A successful backup job does not prove a restore works.
- Test rollback of code and of database changes separately. Switching code back does not undo a destructive migration.
- Errors reach a person. Trigger one controlled alert and confirm who receives it and what they do next.
4. Keep a launch record
For every item, record pass, fail, not tested or not applicable, with the evidence, an owner and a date. Fix what blocks a safe launch first: exposed private data, wrong charges, or a customer locked out of what they paid for. Cosmetic improvements can wait.
Do not turn unchecked items into a “secure” badge. This checklist is not a penetration test or a certification; how much specialist review you need depends on the data you hold and how exposed the product is.
When to ask for help
If you can work through this list yourself, do. If you would rather have an engineer do it with you, our AI-built app production check reviews code, data access, payments and deployment within an agreed scope and returns prioritised findings in 3–5 working days. Fixes are scoped separately. If the app works but cannot grow, the rescue and rebuild sprint is the next step.