I've been using Claude for a bunch of personal tasks as an experiment to see what I can do with it it for my business.
When it works, it is amazing. In using it regularly, however, the unreliability is pretty staggering. One weird thing I've discovered is that it pretends to follow prompts and then will conceal when it screws up. It also will overpromise something it can do and when asked it will say something like, "I'm really sorry I said I could do something but actually, I can't. My bad!"
Overall, it has been interesting but I see no evidence this is legitimately going to change the world anytime soon.