Build on my AI prototype or start over?
The most expensive choice is not the wrong one, it is the postponed one. Five questions that decide whether your project can be saved, and why starting over rarely means throwing work away.
Everyone who comes to me with this question is hoping for the same answer: build on. Understandable, you have weeks in it. Yet that is far from always the cheapest route, and the difference rarely sits where people look for it.
What you keep either way
The most valuable part of your prototype is not the code. It is that you now know which screens you need, which fields belong on them, which exceptions you had forgotten and where your colleagues tripped up. That is design work agencies spend weeks on, and it stays yours even if not a line of code survives.
Which is why “starting over” is a misleading term. You are not starting over, you are building the same thing again with all the answers already in hand. That is a fraction of the work of the first time.
Five questions that make the choice
- Is the data model right? If your tables hold roughly the right things, building on is realistic. If everything sits in one big table with columns like extra_1, it is not.
- How long does a small change take? If a text change takes ten minutes, the project is still alive. If it takes half a day and a broken screen, the end is in sight.
- Is there a conventional stack underneath? Familiar, widely used parts can be taken over. Something exotic the tool invented itself cannot.
- How many screens are there really? Six screens are moved across in no time. Forty screens make rebuilding a serious project all of a sudden.
- Is something already running with real customers on it? If so, building on almost always wins, because standing still costs more than mess does.
The middle road that wins most often
In practice I usually pick neither extreme. What works is keeping the outside and replacing the inside: the screens you have stay, the data model is done properly again, permissions move to the server and the calculations move into one place. From the outside your app looks the same the next day, on the inside it is maintainable.
Build on or start over is rarely the question. The question is which half you throw away.
Why postponing is the most expensive
Every week you build on a foundation that is not right, you add screens that will have to come across too. So the bill grows while you hesitate. And the moment people call is almost always the same: a customer has just come on board who must not notice anything, and now nothing can go down either.
Finding out where you stand costs you nothing. Send me a link or access to the repository and within a few days you will hear which of the two routes I would take, including if the answer is that you can carry on perfectly well yourself.