A super piece, with a lot of really good insights. The "judgement" mentioned is maybe what differentiates between the quick and the dead in business terms. Building the ineffective thing brilliantly and quickly won't move the business' needle.
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.