Four broken projects taught me to prompt one function at a time, not one refactor at a time. The 20-file change came after that shift.

One Saturday to describe it, one Sunday to have it in the store. I kept the prompt template, and the list of what went wrong before that.

Friday idea, Sunday launch with Stripe and auth wired in. A workout planner shipped; the parts I would not build the same way are in here too.

My first AI-built SaaS got seven signups and zero paying users. Two weeks in Cursor, a Product Hunt launch, and a lesson about who decides what to build.

The parser took the AI thirty seconds and cost me six hours on one off-by-one. It cannot run the code, so the assumption was mine.

My first AI app worked and nobody wanted it. Three months on a habit tracker taught me the prompt was never the problem, the target was.