READY qualifies agents for enterprise deployment on reliability and oversight cost, not benchmark accuracy

arXiv·high signalXBlueskyLinkedInCopy link

READY argues an agent can score well on benchmarks and still be undeployable, because enterprises ask whether it hits a reliability target under acceptable human oversight at tolerable cost. The framework takes an agent, a workflow and a class of oversight policies, measures reliability and operating cost of the combined human-AI system, picks the minimum-cost policy meeting the target, and statistically qualifies it on held-out cases. In a clinical-audit study spanning 16 agent systems and 750 cases, two systems separated by 0.3 points of autonomous accuracy (72.8% vs 73.1%) diverged sharply once oversight burden was priced in. The testbed is open and runs on existing agent-evaluation infrastructure.