AI evaluation is moving beyond LLM-as-a-judge toward Agent-as-a-Judge, with translation in scope — promising, but still at an ...