ProAgentBench: Evaluating LLM Agents for Proactive Assistance with Real-World Data

TL;DR

ProAgentBench is a new benchmark with real user data and a hierarchical framework for evaluating proactive AI assistants, addressing limitations of synthetic datasets and isolated task focus in existing research.

Contribution

It introduces a hierarchical task framework, a large real-world dataset, and comprehensive evaluations for proactive AI agents in continuous workflows.

Findings

Long-term memory improves prediction accuracy.

Real-world data outperforms synthetic data.

Historical context enhances proactive assistance.

Abstract

Proactive agents that anticipate user intentions without explicit prompts represent a significant evolution in human-AI interaction, promising to reduce cognitive load and streamline workflows. However, existing datasets suffer from two critical deficiencies: (1) reliance on LLM-synthesized data that fails to capture authentic human decision-making patterns, and (2) focus on isolated tasks rather than continuous workflows, missing the pre-assistance behavioral context essential for learning proactive intervention signals. To address these gaps, we introduce ProAgentBench, a rigorous benchmark for proactive agents in working scenarios. Our contributions include: (1) a hierarchical task framework that decomposes proactive assistance into timing prediction and assist content generation; (2) a privacy-compliant dataset with 28,000+ events from 500+ hours of real user sessions, preserving bursty interaction patterns (burstiness B=0.787) absent in synthetic data; and (3) extensive experiments that evaluates LLM- and VLM-based baselines. Numerically, we showed that long-term memory and historical context significantly enhance prediction accuracy, while real-world training data substantially outperforms synthetic alternatives. We release our dataset and code at https://anonymous.4open.science/r/ProAgentBench-6BC0.