Imagine asking someone to tidy a kitchen.
You come back to a beautiful counter. The dishes have been stacked inside a cupboard, still dirty.
The room photographs well. Making dinner is going to be interesting.
That is an invented example, but it captures something useful about instructions. “Tidy” describes a result you can picture. The other person has to supply their own picture.
AI gets the same assignment when we ask for something “professional”, “useful” or “ready”.
I find the more revealing exercise is to finish the sentence: “I will know it is ready when…”
For the kitchen, that might mean the dishes are washed and put away, the counter is clear, and food is stored where it belongs.
For an AI-written meeting summary, it might mean each decision is separated from discussion, every action names its owner, and an unknown deadline is marked as unknown.
You have given the AI something you can both inspect.
Try this on one small task you already have. Ask your AI:
“Before doing this, turn my description into three observable conditions for a good result. Use only the information I have given you. Flag anything that needs a decision from me.”
Read those conditions before you let it continue. Three is a useful starting limit, not a magic number. Add a condition if it catches something important; remove one if it merely creates work.
You may discover that the AI understood your request perfectly. You may also catch the equivalent of a cupboard full of dirty plates while the job is still easy to change.
A few clear conditions can save a surprisingly long conversation about what “tidy” was supposed to mean.
FELIX