The pathology of generative AI is that it too easily allows substantial form without discernible intent. That mistake is harder to make when creating by hand.

Well put by caleb gross, and:

A human can show up to a task with an unclear mental model of what they mean to accomplish, and an AI can generate something anyway. “Write a letter of resignation for me to send to my boss.” “Hmm…I guess that looks good.”

It's not that you can't show up with unclear intent when writing unassisted - it's that writing unassisted forces you to clarify. I find this in programming too.