1 paper · 1 filter
Jianing Zhu, Yeonju Ro, John Robertson +5
Long-lived AI agents are increasingly deployed as persistent operational systems, yet they are still evaluated like freshly initialized models. Day-one benchmarks miss a basic syst…