Fixing the app you built with AI that won't quite work

You’ve got the idea and the code. Now it needs to work.

AI coding tools like Cursor and Claude Code can get an idea onto a screen in an afternoon. You describe a booking page and it appears. You ask for a customer login and there it is. For a small business owner who’s never written code, that’s a real step forward.

The hard part comes later. The demo looked right, but the buttons do nothing. The login works on your laptop and fails once it’s online. Or you ask the AI to fix one error and it quietly breaks something that worked yesterday.

Find the cause, not the symptom

An error on screen often starts somewhere else. A blank page of bookings might come from a failed login, a database rule, or a missing link to the server. Asking the AI to change the page again won’t fix any of those.

So we start by running your app and making the problem happen. Then we follow it back through the code and the tools it connects to until we find the cause. We explain what we found in plain words, so you know what’s wrong and why.

We also look for the problems you haven’t hit yet. AI coding tools often paste secret keys into the code, skip checks on what customers type in, or let any logged-in customer see everyone’s records. Those need fixing before real customers use the app.

Agree what the next release needs

We work out with you the few things your app must do well. For a booking app, that might be a customer signs in, picks a time that’s free, and gets a booking you can see. That gives the fix a clear finish line.

Some apps need a few focused repairs. Others need part of the app reorganised before any more changes will stick. If starting again is the better choice, we’ll say so, explain why, and tell you what can carry over. You see the plan before we start.

Fix, test, and get it ready

We work on a copy of your app, so nothing you have today is lost. We fix one thing at a time and test each fix against the steps your customers take. We also test what happens when things go wrong, like a customer leaving a field blank or a payment failing.

Then we get it live. We check the setup, put secret keys somewhere safe, and write down how to release a new version and how to undo one that goes wrong.

The same thing happens with AI helpers, not only apps. We work with a three-person online skincare store whose first AI helper sent simple tracking replies on its own. It answered a customer whose jar had arrived broken with a cheerful tracking link. We switched auto-send off, and every reply now waits for the founder. Returns still go through by hand.

Keep control after the fix

A working app is only half the job if you can’t change it next month. We hand over plain notes on how it runs, what we changed, and anything still left to fix. You can keep building with your AI coding tool, now on firmer ground.

If you’d like us to keep looking after it, that’s part of our AI & Technology Partner packages. If the app’s purpose is still fuzzy, start with mapping how you work. We fix apps for small teams across e-commerce, coaching and online education, beauty and wellness, and hospitality and events.