Everything about LLMs is inefficient. They have their benefits but watching them reason over things that are painfully obvious, that they've literally investigated before (before a memory compaction), never take a step back aand be like 'this is going too slow let me look for a better way', etc. is painful.