Read a codebase with clear, consistent, well-established patterns.

One of my concerns with LLM generated code is that it tends to be poorly organized: somewhat the opposite of clear, consistent, well-established patterns. It might replicate well-established patterns on a small scale, but overall the code tends to be poorly organized, repetitious and very often incorrect, requiring another iteration that introduces more disjoint/inconsistent change. Thus, using an LLM to modify code that has been written by good programmers who produce clear, consistent code using well-established patterns works relatively well but using an LLM to modify code produced through many iterations of LLM modification might not work so well. This too changes the economics.