Code is written to be auditable by humans, if you just generate machine code how will you know the level of correctness? Would you save the prompts then I have to prompt review before merge? That would end up being much wordier than a code file.
Code is the spec, and it's far faster to write the code yourself (if you know what you want to do and how to write code) than try to pass all this context to an LLM in the form of lossy and imprecise human language.