It's a tool that you have to babysit, at least for foreseeable future. In general it's always a bad idea for human to supervise the machine because in time we grow complacent of its results and that's when the mistakes happen. When it tomes to writing some content, biggest problem is inaccuracies or some typo. Your comparison to CAD software is not a good one, since CAD doesn't produce anything on its own. It's a software assisting human, not generating content. Imagine the horror with CAD software auto-generated bridges. It would be only a matter of time before someone would just skip on double-checking what was generated. And I am fully aware there are AI generated structural parts and testing, but it's a part of design process where results have to checked by a human again.

I do think AI has a place and purpose, but it's not going to cost people their jobs, only help them do it more efficiently. It's great in assisting people but not replacing. If there's a manager out there who thinks AI can replace a human, then I can show you a bad manager who doesn't understand what AI is. In the future we might arrive at a point in time where AI is good enough to do some jobs human find too repetitive or dangerous. But we are far from that.

Also, LLM is not something I'd call AI, or at least intelligent. They are specialized neural networks which are trained using human input and whose sole purpose is predicting the next word or sentence in relation to what's entered as input. Glorified and overly complicated auto-complete. There's no intelligence involved.