Q&D Training for Horizontal and Vertical Agent Proactivity
September 28, 2026
The Q&D (questioner and drafter) method trains agents to pursue unstated information through horizontal proactivity (contextual) and vertical proactivity (evidence-based). It uses a need graph to score agent behavior without requiring a model judge.
HOW THIS AFFECTS YOU
●
builderYou can develop agents that anticipate user needs and gather necessary context autonomously.
●
researcherThe need graph provides a scalable, model-free evaluation framework for agentic proactivity.