VitaBench 2.0: Evaluating Personalized and Proactive Agents in Long-Term User Interactions

Researchers introduced VitaBench 2.0, a benchmark for testing personalized and proactive AI agents.

It measures whether agents can extract, update, and use fragmented user preferences across ordered interactions.

The benchmark also checks if agents can proactively ask for missing information before making decisions.

The work exposes a gap between current LLM-based agents and real-world personalization needs.

It aims to better evaluate memory, adaptation, and long-term user modeling in AI systems.