Evaluation is becoming a core professional skill: not merely checking whether AI produced something plausible, but whether it improved the outcome.