Resources
Built it with AI and it keeps breaking?
Salvageable or rebuild — 10 questions.
You got further than most people ever do: something that runs, that you can click, that shows what your product should be. Then every new change started breaking something else.
That's a common wall, not a personal failure. This diagnostic tells you which of three paths fits your situation — before you spend another month prompting in circles.
Answer yes or no. If you're not sure, answer no.
Your path is worked out in your browser, which needs JavaScript. Without it, you can still read the 10 questions below.
Read this first
- Back up your data today.
- Plan the data move. Your users' data comes with you.
Your path
Stabilize
Your foundations are usable. What's missing is the safety net.
- Stop adding features for a short while.
- Add tests around your key user flows, so every change proves it didn't break them.
- Clean up what's duplicated or unused, then continue building — carefully, one change at a time.
Typical effort: days to a few weeks.
Partial rebuild
Some of it is worth keeping — usually the screens and the data. The core isn't.
- Keep what works and what users see.
- Rebuild the core (data, business rules, login, payments) properly, with tests.
- Reconnect the screens to the new core, flow by flow.
Typical effort: a few weeks.
Restart from your prototype
The code isn't worth saving. The prototype is.
- It's the best specification you could have written: every screen, every flow, clickable. Most specs are a document; yours is a working demo.
- Use it as the scope for a proper build, with tests from day one.
- Restarting isn't failing. Your prototype proved the idea and showed what the product should be. That was its job.
Typical effort: a full build — a faster one, because the hardest question (what to build) is already answered.
- 01 Does your main user flow work end to end today, for a real user?
- 02 Can you add a small feature without breaking something else?
- 03 Is there at least one automated test?
- 04 Is the code in a repository you control, with its history?
- 06 Could another developer run it on their computer, from written instructions?
- 07 Are login and payments handled by established services — not custom code written from scratch?
- 08 Has your data structure stayed stable over your last few changes?
- 10 When you ask for a change, does it land where you expect — in one place, not ten?
What not to do, whatever the path
- Don't ask the AI to "fix everything" in one go.
- Don't add new features while the app is unstable.
- Don't delete the prototype: it's your specification.
- Don't move real user data without a backup and a test run.
Check your inbox. Click the link in our email to confirm, and it's on its way.
Want a second pair of eyes? An existing build is assessed during our scoping Sprint, before anyone prices the next step. Product Discovery Sprint
Made by NexusInsight. We build with AI too — test-driven. The method makes the difference, not the tool.