No, waiting for "AI" to be more than super-niche useful. I am saying that way too much money has been poured in for the fairly underwhelming results that we get.
Software engineering is a huge industry in which AI has produced enough results to make it all worthwhile.
Image generation is another. I personally know a business owner - it's an online grocery store - who used AI generated pictures to reduce costs by 10s of thousands of Euro, it's what made the business possible. (The pictures are cartoonish, nobody thinks it's a real photo)
I view the time of hand-crafting good code a better investment than spending the same time (and sometimes more) carefully eye-balling tool output that is almost guaranteed to contain subtle mistakes and correct them.
I don't have this problem at all, and yet I use Copilot daily since it was closed beta. The output I get is very helpful, indeed sometimes wrong but it's very much outweighed by the successes. I sometimes turn it off to remember how it used to be, and it's just terrible. It actually makes me and my team 10x devs.
The point is that the senior professionals who benefited from it created so much additional productivity, it doesn't matter that there is someone who can't use it well - that doesn't make AI useless or not worth it.
One anecdotal evidence does not nullify the other, and vice versa. You have your experience, I have mine. I even qualified my statements -- one of the languages I use is not very popular and thus statistical models like the LLMs obviously don't do well with it -- but you are happy to ignore that and keep arguing that your experience is the prevailing phenomena, which I'll always disagree with.
Marginal benefits? Even just Github Copilot itself is absolutely worth it, itself probably a billion dollar company, and changed the world.