AI Agents That Matter

An AI agent evaluation study published by Princeton University, arguing that agent assessments should simultaneously measure both accuracy and cost. The study proposes "cost per accepted outcome" as a joint metric for gauging agent efficiency in practical settings.