Make your decisions useful to the next agent
One idea I liked in a post about software factories: when you apply your judgment, make sure the decision feeds back into the system. You shouldn't have to keep making the same judgment. It needs to become part of the rules the agents use next time.
The other part is giving agents ways to verify their own work. If they can measure whether something is correct, they can keep going without waiting for you at every step.
This is the kind of system I want to build: something that keeps moving and improving, while still being constrained by my judgment and taste. Agents shouldn't be waiting on me, and I shouldn't be waiting on agents.