Assuming the second LLM hadn’t been trained on the existing codebase. Which in this case we can’t know, but can assume that it was.