So… first 80%: zero effort, last 20%, 80% effort. That's a 20% savings right?
Not so fast because picking up where AI left off may make those last 20% take MORE than 100% of the effort. eg. cleaning up an ugly codebase, or an artist trying to recover an AI sketch…
(2/2)

As we get a rational understanding of the areas where AI is good at automating—and where its failure modes are above the risk threshold—we'll converge to a new equilibrium of humans and machines each respectively doing what they are good at.
(3/3)

Less predictability because we're still trying to get a good grasp on failure modes. That's always the interesting aspect in any system. But we're making progress.
(2/3)