Apple introduces Pare framework for evaluating proactive AI agents

Pare models apps as finite state machines for realistic user simulation. The Pare-Bench benchmark includes 143 tasks across communication, productivity, scheduling, and lifestyle apps.