The Prototype Works. That Is the Problem.

The Prototype Works. That Is the Problem.

November 16, 2026

A founder came to us with a product he built himself. He is an expert in his field, not a software developer. Over about two years he built it with AI tools. In his own words, it is mostly vibe coded. It works, and people can use it.

He did not come to us because it is broken. He came because he cannot step away from it.

Three things he told us

It is hard to keep consistent. Each change solves the task in front of it, in its own way. After two years the same kind of problem is solved in several ways across the codebase, and new changes drift away from the old ones.

Nobody knows how it scales. It handles the users it has today. Whether it handles ten times as many is a question nobody has put to the code yet.

The tests prove less than they appear to. The AI wrote the features, and the AI wrote the tests that say the features work. That is code checked against its own understanding. If the understanding was wrong, the feature and the test are wrong together, and the test still passes.

The real cost is his attention

None of those three is why he called. He called because he is stuck between two jobs. If he keeps building, he has no time to market and sell. If he goes out to sell, he has no time to make sure the product holds for the customers he brings in.

The prototype did its job. It proved the idea. Now it is the thing that keeps him from the next step.

What this kind of product needs first

We are still in talks, so this is not a case study. It is how we approach a product in this position.

  1. Read all of it. An engineer reads the codebase end to end and writes down what will break, in order of how much it matters. After two years of generated changes, that is often the first time anyone has read the whole thing.
  2. Write tests from what the product should do. Not from what the code does today. A test written by someone who did not write the feature is the first independent check the product gets.
  3. Pick one way to do each thing. Then write it down where the AI tools read it, so the next generated change follows the same pattern.
  4. Keep the product running. No rewrite. The fixes go into the code that exists, while customers keep using it.

Building it this way was the right call

He built a working product in his own field without hiring a team first. That is a good trade. The mistake would be to treat the prototype as finished, or to throw it away and start over. It needs a different kind of work now: less generating, more reading.

If this sounds like your product, here is how we take a prototype to production. The first look is free.

Google Cloud Partner

Tell Us About Your Project