@aleen so, been thinking about this.

First off, yes.

But (setting aside the sheer qualitative shiftiness of agent-generated prose) I think the problem is at least partly that it’s generated code which is really hard to justify from first principles. Trying to write docs for that code is pretty daunting because it doesn’t have a coherent explanation beyond “this is what made the tests pass.”

But on the other other hand, that could be said about most human-generated code I’ve encountered as well.

Feeling pretty down about it all right now.

But I’ll always be in favour of skilled writers (and skillful writing).