With how verbose LLMs are and how much code churn they generate, it will be very difficult to extract anything and apply it back upstream