You missed the part about unreliable results. Never in software engineering have we had to put a lot of effort into a tool that gives unpredictable, unreliable results like LLMs.