I am probably going to piss off both camps with how I use AI.
For pure writing, I do use Hemingway. I think of it as a souped-up Grammarly. It catches the sentences that got away from me, points at unnecessary words, and makes me look again at passages that are harder to read than I intended. If I am going to sound dumb, I want it to be because of my ideas, not my grammar.
I don’t accept every correction. Sometimes the awkward sentence sounds like me, or the repeated word is there because I wanted to repeat it. I leave those alone.
I am comfortable letting an agent question a draft for the same reason. It can tell me where I lost the thread, where I made the same point twice, or where I assumed the reader knew something I never said. I still have to decide whether it found a problem, and I especially have to decide whether its solution belongs in my voice. Usually it does not, at least not without me pulling it apart first.
Programming is where this gets inconsistent.
When I am purely programming, I do not even use autocomplete. I cannot stand the constant interruption. I start typing a thought and the editor throws a paragraph across it, and now I am not thinking about the code I was going to write - I am reading the code it guessed I might write and deciding whether to accept it. With ADHD, that tiny negotiation every few seconds is exhausting. Turn it off and the editor becomes quiet again. I can hold the problem long enough to solve it.
At the same time, I will hand a coding agent an entire bounded piece of work and let it go. Which probably sounds backwards to both groups. Why reject a three-line completion and accept hundreds of generated lines?
Because they are different activities for me. When I am programming, I want to stay inside the thought without another voice finishing every sentence. When I delegate to an agent, I am defining the work, reviewing evidence, testing the result, and deciding what lands. I can inspect that work when I’m ready, without suggestions appearing while I’m trying to write something myself.
Writers may decide that any machine editing prose makes the prose less mine. Programmers may decide that directing and reviewing generated code is not programming. Fine. I am not trying to win either definition.
The purists can sort out what to call it.