A startup take-home exercise

How far can I get toward something startup-shaped with just a blank folder, claude/codex, and the spark of an idea?

I have worked as a founder, software engineer, and product manager. I build real systems and care deeply about architecture, reliability, security, and maintainability. This is not how I would actually ship production software but is instead an intentional exercise to help test the outer limits of working with AI agents.

The closest analogies are sight-reading music and take-home exams. They require different skills from preparing for a concert or writing a semester-long thesis, respectively, and offer a way to see what is possible under constraints.

Quick judgment and taste play an outsized role here, and any edges show. What should a serious builder be able to do in one day? That is the bar I’m setting for myself.