The Problem

Building got fast. Knowing whether it mattered got harder.

I am a solo founder on a limited budget. The job is to find which ideas show traction with real customers, as quickly and as cheaply as I can, and nothing does that like a working product in front of someone. Every project stays on free infrastructure until it earns something, which is a rule I hold myself to rather than a preference.

Claude and Codex made building genuinely fast, and I assumed that was the answer. What actually happened is that the difficulty moved somewhere I was not looking.

1 > I could build faster than I could judge

Output stopped being the constraint, and review quietly became one. On a good day I read the code properly, checked it on a phone and rewrote the copy that sounded machine-written. On a bad day I shipped it. That is not a standard, it is however much attention I had left that evening.

The cost: work went out that I would not have approved on a better day.

2 > I could not tell whether any of it mattered

One of my sites took real card payments for six weeks. In that time the work went into the journal, the social posts, the SEO. When I finally sat down to check, nothing I had written down could tell me whether it had sold a single thing.

The cost: six weeks is survivable. Not knowing which of those weeks was worth repeating is not.

3 > Nothing I learned in one project reached the others

I run several projects at once, at different stages. Each had its own copy of the team, and a copy is correct on the day you make it and silently wrong from then on. One was still following a process I had retired weeks earlier, and nothing told me. It failed in the other direction too: something learned on a Tuesday went into whichever copy happened to be open, and disappeared the next time that team was rebuilt.

The cost: five projects, five teams, and not one of them getting better because of the others.

None of these are AI problems

They are the ordinary problems a product organisation solves with people, process and institutional memory. A company handles them with a product manager who holds the line on quality, an analyst who insists on knowing whether it worked, and enough shared practice that nobody relearns the same lesson twice.

I had none of that, and could not afford to hire any of it. So I built it instead, and I made it open in case someone in the same position finds it useful.

If any word on this site is doing work you cannot see, the glossary defines each of them in a sentence.

See the solution →