As a PM, I am starting to see similar dynamics in my work as I use AI more and more. As a PM, my speed of experimentation and bets I can take has gone up. But when I have shallow understanding of the system I am optimizing, the misalignment is amplified. As you said, the work has shifted in clarifying intent - which comes from having deeper context and a point of view. The real cost is spending experimentation cycles on something I don’t fully understand