You built an app with an AI tool. It runs. You’ve clicked through every screen and it feels great. Maybe you’ve even shown it to a few people.
Here’s the uncomfortable part: that doesn’t mean it’s ready for real customers.
AI app builders like Bolt, Lovable and Cursor are genuinely impressive. We use AI tools every day ourselves. But they’re built to get you to “it works when I try it” as fast as possible. A working demo and a working product are two different things, and the gap between them is where businesses get hurt.
The demo trap
When you test your own app, you do everything right. You type sensible things into the boxes. You click each button once. You’re the only person using it.
Real users don’t behave like that. They paste in emails with trailing spaces. They double-click “Submit.” They open the app on bad hotel Wi-Fi. A few of them, sooner or later, will poke around on purpose.
An app that has never met that kind of traffic isn’t tested. It’s just untried.
A real example
We recently worked on an app that had been built entirely in Bolt. On the surface it was polished: nice screens, everything clickable, happy owner.
Underneath, it had effectively no security, and almost no checks on what users typed in. Anyone who knew where to look could see or change other people’s data. Forms accepted whatever they were handed, including garbage that would break things later. None of it was visible from the front, because nothing had ever gone wrong yet.
Nobody did anything wrong here. The owner asked the AI for an app, and got one. They just never knew to ask for the parts you can’t see.
What’s usually missing
Here’s the checklist we run through on almost every AI-built app.
1. Login and permissions. Can user A see user B’s data? Many apps check that you’re logged in, but not whether the data you’re asking for is yours.
2. Secrets. API keys and passwords should live somewhere private. In AI-built apps they often sit in the code itself, or get sent to every visitor’s browser, where anyone can copy them.
3. Input checking. Every box a user can type in is a door. Without validation, bad data gets in, and so can worse.
4. Error handling. When something fails, does the user see a helpful message, or a blank screen? Does anyone find out it failed?
5. Backups. If your database disappeared tonight, could you get it back? “I think the platform handles that” is not a yes.
6. Monitoring. Would you know the app broke before a customer told you?
7. Cost controls. If your app calls an AI service, one bug or one bad actor can run up a bill with no ceiling. A usage limit is cheap insurance.
Why AI tools skip these
It isn’t a flaw so much as a blind spot. AI builds what you ask for. You asked for a booking page, so you got a booking page. Nobody types “and make sure users can’t read each other’s bookings, and back it up nightly, and tell me when it crashes.”
If you don’t know these things exist, you can’t ask for them. That’s the real gap, and it’s exactly what an experienced developer brings.
A 5-minute self-check
Answer these without opening any code:
- Can you log in as one user and confirm you can’t see another user’s data?
- Do you know where your passwords and API keys are stored, and that they’re not in the app’s visible code?
- Do your forms reject obviously bad input (empty fields, wrong formats, huge text)?
- If something fails, does the user see a clear message?
- Do you have a backup you’ve actually restored at least once?
- Would you get an alert if the app went down?
- Is there a cap on what your AI or API usage can cost per day?
Scoring: Mostly yes? You’re in good shape. Three or fewer? Get it looked at before you launch, or soon if you already have.
What to do next
The good news: you almost never need to throw it away. In our experience, most AI-built apps are 70–80% of the way there. The screens, the flow and the idea are real and worth keeping. What’s missing is the foundation underneath.
Fixing that is far cheaper than a rebuild, and far cheaper than a breach.
Built something with AI and not sure it’s ready? Send us a note and we’ll take a look. We’ll tell you plainly what’s solid, what isn’t, and what it would take to fix.
Want the checklist on one page? Download the PDF.