Instruction: Explain how you would keep human review useful without making the system unusably slow.
Context: Tests how the candidate diagnoses the problem, chooses the safest next step, and reasons through recovery. Explain how you would keep human review useful without making the system unusably slow.
Official answer available
Preview the opening of the answer, then unlock the full walkthrough.
I would redesign the approval policy around risk, not around habit. If reviewers are approving nearly everything, the system is probably asking for human input too often or at the wrong granularity.
I would group low-risk actions into broader approval boundaries, automate obviously safe steps with deterministic checks,...
easy
easy
easy
easy
easy
easy