His advice really is 'good user experience is good; bad user experience is bad.' Human intervention is not essential to good user experience in most of his examples; he compares the human versions to poor implementations of automation. A form with some validation and normalization is probably a better user experience than having a human review the fields.
I wouldn't say that was his advice at all. I read it as being something more like "there isn't always a (known, or sensible-ROI to research) way to make an automated user experience anywhere near as good as what you get by just having a human do it."
