Exactly. When you boil it all down judgement is THE bottle neck when it comes to deploying agents at scale to run a real business. The other thing I've learned is that knowing both how to remove judgement from a project/task and also when to do so is a step increase in how much productivity gains you can get from agentic systems. Coding/dev has already largely solved this (loops, /goal, etc) but for real world tasks it's a lot more complicated and the solutions aren't that readily available.