Foundations of Proactive Agents: Principles, Technical Layers, and Proactivity-Gym
A paper defines 3T principles and PROACTIVITY-GYM for evaluating proactive LLM agents on capability, timing, and trust.
The paper proposes three joint principles for proactive LLM agents: Task Capability, Temporal Allocation, and Trust. It defines a design space covering task scope, anticipation horizon, activation trigger, processing timing, and intervention depth, and introduces PROACTIVITY-GYM, a multi-day simulation with persona-conditioned users. Evaluations of 23 model-harness configurations find large gaps and that LLM judges often conflate capability with trust. A 30-participant study shows trust falls after misaligned interventions even when outcomes are correct, and users prefer sleep-time assistance.
- 3T principles cover task capability, compute timing, and user trust.
- PROACTIVITY-GYM simulates multi-day stateful proactive assistance.
- Twenty-three configurations show judges conflate task success with trust.
- Thirty participants prefer imperfect sleep-time help over interrupting work.
From the original article
1 min read at huggingface.co
Proactive LLM agents can turn idle compute into useful support before users ask. Yet even correct work can misread user context, impose review costs, or undermine trust.
By Jio Oh, Seunghyun Do, Young-Jun Lee · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.