But you can't successfully outsource all of "thinking" to a LLM today(or, likely, ever), so this is mostly a strawman.

Thinking isn't Boolean either. A LLM let's you not think about, say, rust lifetimes, as an example. You'll still need to think about other things.

I'm not abandoning all thinking, I just change what things I think about. Some things I delegate thinking about(like rust lifetime syntax, to stick with my example), which frees up time to think about other, more interesting things, like application design, or architecture!