This is spot-on but I feel there is a force to be reckoned with there. How much control is a human willing to cede to the machine? I've noticed myself ceding quite a bit, but starting to feel a limit. There are certainly times when I wished Claude/Codex/etc. wouldn't make so many decisions itself and ask me instead. That leads to friction and redo.
The higher the abstraction level, the more control you cede and the more decisions the machine makes. Maybe that's solved in control theory, I don't know.
Yes agree, and today people use the "interview" and planning skill as a method for the model to learn what your preferences are. I think as the model gets better at that, it can get better at that (both generally and in context of a repo or individual env). But also, if it presents you with two fully implemented branches and you just choose which one you like better, I think that would that's also a good experience. You're not asking for a redo, you're just picking one.
This is spot-on but I feel there is a force to be reckoned with there. How much control is a human willing to cede to the machine? I've noticed myself ceding quite a bit, but starting to feel a limit. There are certainly times when I wished Claude/Codex/etc. wouldn't make so many decisions itself and ask me instead. That leads to friction and redo.
The higher the abstraction level, the more control you cede and the more decisions the machine makes. Maybe that's solved in control theory, I don't know.
Yes agree, and today people use the "interview" and planning skill as a method for the model to learn what your preferences are. I think as the model gets better at that, it can get better at that (both generally and in context of a repo or individual env). But also, if it presents you with two fully implemented branches and you just choose which one you like better, I think that would that's also a good experience. You're not asking for a redo, you're just picking one.