https://github.com/arpentry/arpentry/blob/main/docs/MOTIVATION.md

I wanted to test a question: if producing code costs almost nothing, how would you build a globe viewer? Instead of optimizing for human productivity (high-level language, existing engine, standard formats), what if you optimize for the output: binary size, portability, runtime cost, control?

So I let Claude produce it in C targeting WebGPU. Documentation first, code second. Given clear specs, it produced working C on the first pass. Given vague specs, the code wandered or chose the obvious path. This is experimental code and there are many areas for improvement, but I was pleasantly surprised at the results.

Repository: https://github.com/arpentry/arpentry