I am also at the borderline of not wanting to do AI anymore or at least not at the circumstances of the AI projects I have been on. However, for a totally different reason. In total opposition of what you said, I think the stupidest thing at the moment is trying to go towards AGI or this "fancy movie AI" stereotype. Instead, what the world really needs right now are systems that can do really specific stuff really well.

However, when people think of AI they forget that >it might not work< or that >it might need way more effort than expected<. Besides, they try to stick AI development to practices such as SCRUM, as if you could reliably tell how long you will take to make a model good.

I don't know if I have just been on the wrong projects or if that's a real widespread issue, but, for now, I long for more old school development, that you regularly ship stuff without much of a doubt whether it will work or not.