I would rather the junior go do that in their own time and get back to me when they have figured it all out. I don't want to babysit juniors, I want to mentor them and then give them the lead and time to figure out the minutiae. That gives me time to get stuff done too. With AI right now, you end up down a senior while they are babysitting a rapid junior.

I have found it useful for starting novel tasks by seeing if there's already an established approach, but I pretty well always have to fudge it into a real application which is again, the part I want the junior to do.

That's like comparing a mathematician to a calculator. The LLM won't do anything useful if you aren't providing it with a perpetual sequence of instructions.

Sure, but unlike a junior dev its output wont improve with time. You'll have to hold its hand for eternity.