Published Engineering
How to Tell If an AI-Built App Is Safe to Launch
You built it with Lovable, Cursor or Claude and it looks right. Here is what safe to launch means and six questions to ask before real users arrive.
You built an app with an AI tool. Maybe you started in Lovable and moved to Claude. It looks right and it works when you click through it. But you can't see what's going on underneath, and you're about to put real users on it.
This is what I check when someone asks me whether an AI-built app is safe to launch, and the questions you can ask yourself or your developer first.
What "safe to launch" means
Three things:
- Users can be on it. People can sign up, sign in and use it without it falling over.
- Their data is protected. Nobody can see or change what they shouldn't, and there's no breach waiting to happen.
- It stands up. It keeps running, and when something does break, you find out.
A working demo tells you none of these three. AI tools are very good at making something look finished. They are much less careful about the parts you can't see.
Six questions to ask
1. How is it hosted, and how is that set up? Where does the app run, who has access to that account, and how is it configured? This is the first thing I look at, because everything else sits on top of it.
2. Can one user read another user's data? If your app stores data in a database, there should be rules saying which person can read which rows. They're often called row level security, or RLS. When they're missing, anybody who knows where to look can read other people's data. This is one of the first things I check, and one of the worst to get wrong.
3. Does everyone only get the access they should? This is authorisation. Can an ordinary user do something only an admin should? Can someone see another account just by changing a number in the address? If the permissions are loose, people end up with more access than you meant to give them.
4. Are there passwords or API keys in the code? Keys sitting in the code can be found and used by other people. Check how the app handles secrets, and how it moves between development, testing and live, because that's where they tend to leak.
5. If something breaks, will anyone know? Some apps have no logging at all. Others only print messages that nobody ever sees. Ask what happens when something fails. Is it recorded? Is it retried? Does anyone get an alert, and does the app tell the user? If the answer is that nothing happens, then failures are silent, and you'll hear about them from your customers.
6. Do sign up, sign in and onboarding actually work properly? User accounts are where I see the biggest problems. Check that signing up and signing in work properly, that each person ends up with the right permissions, and that the first experience makes sense. A bad onboarding is a sign the basics weren't thought through.
Signs it isn't ready
Everything above counts. The ones you can spot without reading any code are:
- Accounts and permissions that don't behave properly.
- A confusing first experience.
- No one can tell you what happens when something goes wrong.
If your developer or your AI tool can't answer questions 2 to 5 clearly, treat that as a no for now.
When a scanner is enough
An automated scanner is a good first step. It's quick, and it's good at things like secrets and sensitive data left in the code. Run one.
But a scanner doesn't judge structure. It can't tell you whether the app is built in a sensible way, whether the pieces talk to each other safely, or whether the code is spaghetti that will break the first time you change something. That takes someone reading it.
What an independent review gives you
When I review an AI-built app, I go in this order: the hosting and its configuration, then the architecture and data, including security, RLS, authentication and authorisation, and how the parts communicate. Next, whether you can see what's happening, so logging, failure and retry behaviour, and alerting. After that the environments, deployment and secrets, and finally the code itself: sensitive data leaking, how it's organised, and what will be painful to change.
You get a written answer on whether it's ready, and a list of what to fix first. There's more on this on the code audit for software built with AI tools page. If you aren't sure where your app stands, get in touch and tell me what you've built.
Related posts
- 14 Nov 2026Code Audit vs Penetration Test: Which Do You Need?A code audit reads your software from the inside. A penetration test attacks it from the outside. What each tells you, and which to buy first.
- 10 Oct 2026Can You Trust a Platform Your Supplier Built?What you can check yourself about software an agency or freelancer built, what needs an independent view, and who owns the code, repositories and hosting.
- 10 Oct 2026Questions to Ask Before You Commission a Technical ReviewA checklist for non-technical buyers: who does the work, what access they need, what you receive, and what happens if the result is bad.