I found those tools to resemble an intern: they can do some tasks pretty well, when explained just right, but others you'd spend more time guiding than it would have taken you to do it yourself.
And rarely can you or the model/intern can tell ahead of time which tasks are in each of those categories.
The difference is, interns grow and become useful in months: the current rate of improvements in those tools isn't even close to that of most interns.
I have a slightly different view. IMHO LLMs are excellent rubber ducks or pair programmers. The rate at which I can try ideas and get them back is much higher than what I would be doing by myself. It gets me unstuck in places where I might have spent the best part of a day in the past.
My experience differs: if at all, they get me unstuck by trying to shove bad ideas, which allows me to realize "oh, that's bad, let's not do that". But it's also extremely frustrating, because a stream of bad ideas from a human has some hope they'll learn, but here I know I'll get the same BS, only with an annoying and inhumane apology boilerplate.
And rarely can you or the model/intern can tell ahead of time which tasks are in each of those categories.
The difference is, interns grow and become useful in months: the current rate of improvements in those tools isn't even close to that of most interns.