Apple Proposes Pare: A Proactive Agent Evaluation Environment

Apple ML Research · rss · 2026-07-14

Apple ML Research introduced Proactive Agent Research Environment (Pare) for building and evaluating proactive agents. The paper notes that current practices simplify applications into flat tool-calling APIs, failing to capture the stateful, sequential nature of real user interactions, making credible user simulation difficult.

Pare models applications as finite state machines to simulate proactive user behavior in digital environments and evaluate proactive assistants accordingly. The goal is to address the lack of realistic user simulation, providing a framework for training and evaluating proactive assistants that closely mirrors actual interactions.

Original post →

More from Apps

Apps channel →