how good are LLMs at maintaining a consistent code style or "voice"? things like context-dependent stylistic decisions that you want to enforce, in addition to general rules that should be easier to lint.

I ask because after iterating with AI for sometime, the code starts to feel like unmolded dough because of all the stylistic choices it makes. do people just swoop in at the end and re-shape the code, or do they enforce it somehow as they go?

the idea is to work quickly to produce working code while reliably producing finalized output that is human-readable in the same way I'd want to write it. thanks!