The "write prompt first, measure both approaches" experiment is genius. Eliminates the bias of learning from doing the task yourself. Stealing this methodology! Also love the paralegal analogy - captures the current state of AI agents perfectly.

I'm happy you work in C. I hope to be working in C too. I mean you keep making the $ through C, that's really nice