Manual testing, and making sure that the AI didn't create so many bugs.
But, to the underlying question, obviously as we automate more and more of our work, of course we provide less and less value. We're heading towards a future where selling thought for money isn't going to work so well.
