Five signs your AI prototype is hitting its ceiling
Lovable, Bolt and v0 get you surprisingly far. Right up to the point where every change breaks something else. How to spot that moment before it costs you the project.
I have seen it often enough by now to recognise the pattern. Somebody builds something over a weekend that looks good and genuinely works, shows it to customers, gets enthusiastic responses, and grinds to a complete halt three months later. Not because the tools are bad, but because a prototype has different requirements than a product.
1. Every change breaks something else
You ask for a tweak on one screen and something breaks on a screen you have not touched today. That is not coincidence: the same logic is copied in five places and there are no tests to warn you. How long small changes take becomes the clearest gauge.
2. You no longer dare delete data
There are rows in your database and nobody knows any more whether they are in use. There is no migration history, so you cannot trace when a column arrived or why. This is the point at which you start staying away from your own software, and that is the beginning of the end.
3. Security was never set up explicitly
You can see what you are allowed to see because the screen does not show it, not because the server refuses. As long as you are the only one logging in, that goes unnoticed. The moment real users with real data are in there, this is where it goes from annoying to legally reportable.
4. There is no separation between testing and real
One environment, so you are testing on your real customers. Every time you try something, a user can see it. At some point you only dare work in the evenings, and that is no way to maintain software.
5. Nobody can explain how it works
You have thousands of lines of code you did not write and have not read. As long as everything works, that is not a problem. At the first outage it is a big problem, because you cannot search for something whose structure you do not know.
A prototype that customers use is no longer a prototype. It is production software without the arrangements that come with it.
What to do then
Not throw it all away. What you worked out about your own process in that prototype is valuable and has saved you months of design work. What you carry over is the knowledge, not necessarily the code: the data model done properly again, authorisation on the server, a separate test environment and tests in the places where money or data move. In practice that costs less than starting over, precisely because the questions have already been answered.