AI
Asking for What Was Never Requested: Horizontal and Vertical Proactivity in Agents
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 28/09/2026, 20:00:00 UTC
Briefing created: 30/09/2026, 09:47:32 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 09:18:24 UTC.
Briefings
This article investigates horizontal and vertical proactivity in AI agents, defined as pursuing unstated or earlier-revealed information respectively. Researchers propose Q&D, a method training agents to select questions that maximize evidence retrieval without external reward models. On three multi-hop benchmarks, the trained questioner improved both proactivity types compared to standard prompting and outperformed a 15x larger model on two tasks. In a simulated customer service scenario, the agent completed more tasks while asking fewer questions, and in retail settings, it outperformed the larger model with fewer customer follow-up turns. The study concludes that proactivity depends on both the agent's decision to act and its strategic choice of pursuit.
Key points
- Horizontal proactivity pursues unstated information identified by the current context, while vertical proactivity pursues needs revealed only by earlier evidence.
- The Q&D method trains agents to select questions that maximize evidence retrieval without using reward models or human judges.
Context and limitations
Results are based on held-out splits of three multi-hop question-answering benchmarks and a simulated customer service scenario; no model judge was used for scoring proactivity.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/09/2026, 09:47:32 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeichge67wylq4e7ac4gggd66rx7xbzbwkpeaguosek3qrmofqvttqy