It might not be obvious based on a lot of my recent tweets, but we use AI to write a lot of code at Bootdev (but not course content, and we do read every line of code)
But yeah, most of the platform's code is at some point coming out of an agent.
That said, we recently had to ban using agents to write issues. Sure, we use agents to pull context, do research, etc. But it's just SO GODDAMN ANNOYING when you are trying to read and interpret a spec that someone else couldn't even be bothered to take the time to write.
For code, at the line level, agents write very reasonably. And when you know enough to critique their work, you can get it exactly how you want it.
For prose, at the line level, AI is absolute dogwater, and no amount of prompting can fix it with the current models. And at this point, if you disagree, I believe you're just a bad writer.