I'm firmly in the "programming is fun" camp, I enjoy thinking about the problems and solutions, but if I can get a LLM to deal with boilerplate then that's all the better for me.
1. AI agents aren't just typists. They'll fill in the blanks and implement something that works from extremely vague instructions and probably get it right too, because you and your users are predictable. You could be strict and throw out any changes that weren't explicitly what you asked for, but it becomes tiring and it will shape the way you think anyway. It becomes extremely tempting to increase the boundary of what you consider "boilerplate",
2. You probably have colleagues who don't like programming but please the managers because, of course, everything needs to get done right now, as soon as possible. Maybe one day we'll get organisations where enjoyment and understanding matters, but right now we've got tons of people who got in it for the money, just like to "build stuff" and don't care one iota about understanding (and they're oddly proud of this fact),
3. I'm starting to think the typing was a necessary limitation on output speed. I just don't think anyone can think as fast as LLMs can churn out code. You can easily get addicted to the speed and forgo understanding. Like point 1 this is insidious. If you're not really careful you just won't be able to keep up and before long you too will have a bunch of code you don't understand.