We UC
Back to blog
·7 min read·We UC Team

My AI-built app breaks every time I add a feature: why, and who can fix it?

Why a change in one place breaks another in apps built with Lovable, Bolt, Cursor or Replit, what to check first, and how a UK team fixes it without starting again.

My AI-built app breaks every time I add a feature: why, and who can fix it?

An AI-built app usually breaks when you add a feature because the same piece of logic has been written in several places, nothing automatically checks that the old behaviour still works, and the data underneath has a shape the new code doesn't expect. It is a common stage in the life of an app, not a sign that you did anything wrong, and in most cases the app can be steadied without being rebuilt. At We UC we do that with a fixed-price Rescue audit, a written report that shows where the tangles are, and then a choice between fixing, rebuilding the parts that need it, or taking the plan to someone else.

Why does one change break something that was working?

One change breaks another in an AI-built app because the parts depend on each other in ways nobody wrote down. When an AI tool adds a feature, it reads enough of the project to make the new thing work, and it may copy a rule such as "who can see this order" into the new screen instead of reusing the one that already exists. Now there are two copies. Change one and the other keeps the old behaviour, or the new screen assumes a column in the database that an earlier change renamed. The app has no automated tests, which are small programs that run the important journeys (sign up, pay, send the invoice) after every change, so the first person to notice is a customer.

None of this needs bad intent or poor tools. Lovable, Bolt, Cursor, Replit and Claude are very good at getting something working in front of users, and the stage after that, where the app has to be changed safely by one person or many, is a different kind of work.

What should I check before I ask for another feature?

Check four things this week: whether you can go back to a known-good version, whether the code lives somewhere you control, whether there is a copy of your data, and whether the app has a second environment to try changes in. Each is a question you can put to whoever is building it and expect a plain answer.

The first deserves a warning. Lovable's own version history documentation says that "reverting restores your project's code only" and that it "does not restore or roll back your database data." If a bad change altered the structure or contents of your database, pressing revert fixes the screens and leaves the data as the bad change left it. So the copy of your data matters more than the version history does, and it should be a backup you have actually tested restoring.

For the second, the code should sit in a repository (a folder of code with a full history of changes) on an account in your name, for example on GitHub. For the fourth, a staging copy is a second, private version of the app where a change can be tried on dummy data before real customers meet it.

Is it a security problem as well as a reliability problem?

Yes, it can be a security problem as well as a reliability one, because a change that breaks the app's behaviour can also change who is allowed to see what. Many AI-built apps keep their data in Supabase, and Supabase's row level security documentation states that "a table in an exposed schema without RLS is readable and writable by any role with a grant on it." Row level security is a rule inside the database that says a customer may read their own rows and nobody else's. If a new feature adds a table and the rule isn't added with it, that table can be open. Our guide on how to tell whether a vibe-coded app is secure lists five questions for exactly this.

Do I have to rebuild the app?

You usually do not have to rebuild an AI-built app, because a written review settles whether to fix or rebuild and usually a lot more can be kept than people fear. Screens, the data model and the integrations that work are worth keeping, and the pieces that are duplicated or fragile are the ones that get rebuilt. Our article on fixing or rebuilding a vibe-coded app sets out how that call is made.

Who can help me fix it properly in the UK?

A UK software company, an agency or an experienced freelancer can all do it, and the thing to look for is whether they start by reading the code and the hosting and writing down what they find, before changing anything. At We UC, our Rescue audit works that way: we need read access to the code and hosting, we change nothing during the audit, and you receive a written report covering security, data, hosting, code, payments, integrations and running costs, followed by a walkthrough call. The price is £2,500 + VAT, fixed, and it is deducted from the first phase if you go ahead with us on the fix. The report is yours either way, so you can hand it to another team. If you are weighing the options, our piece on choosing a freelancer or an agency gives five questions to ask.

The fix itself is done in fixed-price phases, urgent items first. Two QA engineers test work before it reaches customers, and an engineer reviews every change before it ships. Our main engineering team is in Southampton, with some engineers offshore in Ukraine and Eastern Europe. We run Olatti, Quenta and Scheduler in production ourselves, so the review and QA habits described here are ones we depend on.

What if a consultant built it and has moved on?

If a consultant built the app and has moved on, the steps are the same as for any rescue, with one addition: find out which accounts the app runs on and whether any are in the consultant's name. Hosting, database, repository, domain and payment provider are the ones that cannot be worked around. Our article on what to do when the developer has left goes through what to ask for and how code ownership works in the UK.

Common questions

Will adding features keep breaking things forever?

Not if the causes are dealt with. Repeated logic can be gathered into one place, tests can be added around the journeys that make money, and a staging copy means changes are tried before customers see them. How long that takes depends on the app, and we would size it in the audit.

Can I keep using Lovable, Bolt or Cursor afterwards?

Often yes, and we use AI tools heavily ourselves for speed. The difference is that an engineer reviews every change and QA engineers test the work before it reaches customers, and in a rescue we would look at adding tests around the journeys that make money.

How much does it cost to find out what is wrong?

The Rescue audit is £2,500 + VAT, fixed, and the fee comes off the first phase of any fix we do. Our article on what a code audit costs for an AI-built app explains what that covers.

If your app has real users or money passing through it and each new feature is a gamble, you can book a Rescue audit, or tell us what you have and we'll say whether an audit is the right first step.