The discussion we were having is the extent to which the
Feist test that is mentioned in the PDF applies. Especially considering pages 24-27 in the PDF.
It makes it seem that AI coding assistants are totally fine and the whole work is copyrightable as long as the human is bringing the code pieces in, deciding where they fit, etc. But if you are making "small" edits to a work that is primarily AI-generated then you are likely out of luck. I guess it really depends on how much you have to edit the AI work.
A few days ago on HN there was talk about vibe-coding in the current YC cohort in which they were saying "don't bother debugging, throw it all away and have the AI regenerate!" I can't help but think that this is incredibly risky. One disgruntled ex-employee blows the whistle and your startup might be toast.