All posts

Built it with AI? A 10-point checklist before real customers log in

Lovable, Bolt, Replit and Cursor make it easy to get an app working. Here's what to check before paying customers trust it with their data — most of it you can do yourself.

AI app builders have changed who gets to make software. A founder with a clear idea can now have something clickable by the weekend, and that's genuinely good news. The catch is that these tools are tuned to make the screen you asked for appear. They're much less interested in the parts nobody sees — and those are exactly the parts that matter once strangers start typing their personal details into your app.

This checklist is for the moment just before that happens. It won't make you a security engineer, but it will tell you whether you're ready, nearly ready, or in need of help.

Why "it works" isn't the same as "it's ready"

A prototype has one job: prove that people want the thing. A production app has a different job: protect the people using it, keep running when you're asleep, and survive the next change without breaking.

AI builders are excellent at the first job. They tend to skip the second because nothing in a demo forces the issue. Nobody tries to read someone else's orders during a demo. Nobody deletes the database. Nobody notices the app went down at 3am. Real life does all of those things eventually.

The 10-point checklist

Work through these in order. The first four matter most because they're about other people's data.

  1. Try to see someone else's data. Log in as an ordinary user, open a page showing one of their records, and change the ID in the address bar. If you can see a record that isn't yours, your app checks who is logged in but not what they're allowed to see.
  2. Check your database rules. If you're on Supabase, open the dashboard and look at row-level security for every table. Tables with no policies, or policies that allow everything, need fixing first.
  3. Look for secret keys in the browser. Open developer tools (F12), use the app, and watch the network requests. Payment, email and AI keys that can spend money should only ever live on a server. If one has appeared in the browser, assume it's been seen — rotate it.
  4. List what personal data you hold. Export your data and actually look at it. You may be collecting more than you need, and in Australia the Australian Privacy Principles apply to many businesses that hold customer information.
  5. Restore a backup — for real. Having backups and being able to restore one are two different things. Export your data, then practise restoring it into a copy of the project.
  6. Make sure someone gets told when it breaks. Set up uptime monitoring and error alerts that reach a person, not an inbox nobody reads.
  7. Write down how the database has changed. If the structure has been edited by prompt after prompt, start keeping a record of every change (developers call these migrations). Without it, rebuilding or moving the app later is guesswork.
  8. Add a handful of tests for the things that make money. Sign-up, login, payment, the core action. A few automated tests that run before every change will catch the breakages a new prompt quietly introduces.
  9. Know where it's hosted and who holds the logins. The domain, hosting, database and payment accounts should be in the business's name, with credentials stored somewhere other than one person's laptop.
  10. Check you could leave the platform. If your builder changed its pricing or shut down a feature tomorrow, could you export the code and run it elsewhere?

What your answers mean

Be honest about where you land. Not every prototype needs rescuing.

Where you areWhat to do next
Still testing the idea, no real usersKeep building. Fix nothing yet — it's cheap to throw away.
A few friendly testers, no money changing handsFix points 1–3 and take a backup. Then keep going.
Paying users and personal dataThe foundation needs work before you grow. Points 1–6 are urgent.
An investor or business customer asking about securityEverything above, plus documentation they can read.

Fix it yourself, or get help?

Plenty of founders can work through points 1–6 with a free weekend and some patience. Where it gets harder is when the data model itself has grown messy — duplicate tables, fields that mean two different things, records that no longer line up. Patching permissions on top of that tends to be slower and riskier than rebuilding the foundation underneath while keeping the screens your users already know.

That's what our vibe code cleanup work is: we read the code, tell you plainly whether it needs hardening or a new foundation, and then do it on a live link while your prototype keeps running. Your users keep their accounts and data; what changes is everything they don't see. You can see the kind of production systems we build — with real users, bookings and payments — in our case files for Kaladi Kitchens and Sprout Lessons.

Can you keep using AI tools afterwards?

Yes, and you should. A codebase with tests, a review step and a clear structure is a safer place to use AI coding tools, because mistakes get caught before they reach customers. Hardening the app isn't about giving up the workflow that got you here. It's about giving it guardrails.

Common questions

How long does it take to make an AI-built app production-ready?

It depends on what the app does and the state of the data underneath. A small app with a tidy data model might need a focused hardening pass; one with payments, roles and a drifting database needs more. You'll typically see a working, hardened version within weeks, and we fix the scope and price after reading the code.

Will my users notice anything change?

Usually not, unless you want them to. The screens and flows they already know become the specification. What changes is underneath: permissions, backups, monitoring and reliability.

Do I own the code afterwards?

Yes. The code lives in your repository and runs in accounts in your name. You can host it anywhere and hire anyone to work on it.

What if my app is already leaking data?

Shut off the leak first: fix the database rules or permissions involved, rotate any exposed keys, and work out what was accessed. Then get advice on your notification obligations. If you need a hand, get in touch and say it's urgent.

Want this looked at for your business?

A short call, no pitch. Tell us how things run today and we'll tell you honestly what's worth doing first.