As great as a nicely formed prompt to a llm and its output may be, how much actual productivity are we talking compared with an experienced internet stackoverflow searcher with multitudes of completed projects in their experienced completed code archive to search on how to they did things, or perhaps those thing once known as books. Suddenly we go from the 3x to 4x to 1.5x to 2x factoring in prompt creation time and its tweaking, for ok to great llm output. It's not that llm code is lengthy, its that an expert, one that is professional must know the code to almost the point of memorization to at least total understanding of it.