When incomplete instructions can still be sufficient
An instruction is complete when the interpretations consistent with it and the shared context lead to the same set of actions in each situation. Even when interpretations differ, an action that is optimal for all of them can allow the AI to proceed without clarification.
The cost of specifying and clarifying
The model compares doing the work directly, writing a complete instruction, and delegating with clarification at selected decision points. It accounts for the time spent writing instructions and answering questions, including the value of reusing an answer in later cases.