Parallel agents ask peers only when they need specific help
Ask on Demand: Targeted Peer Communication for Parallel Agentic Systems
Artificial Intelligence
Summary
Solving hard problems by having many AI agents work in parallel often leads to each agent keeping useful clues to itself. The authors show that if agents wait until they actually need some piece of information from a peer and then ask for that specific help, they can work better together. They created a system called Ask on Demand (AOD) that lets agents request targeted advice, choosing the best peer and relevant info for that moment. Tests show this approach improves success rates while using less computing power compared to other ways agents share information.
What this means in practice
- •For ai application developers: Improve multi-agent AI systems to ask targeted questions only when they need specific information from peers.
- •For collaborative robotics teams: Enable robots working in parallel to request precise help from other robots efficiently to complete complex tasks.
Authors
Haofu Yang, Hongyang Yin, Ran Bi, Faguo Wu, Peng Xue, Feng Zhou, Han Zhang
Abstract
Parallel LLM agents tackle complex tasks by exploring multiple trajectories, yet useful discoveries remain isolated in individual execution histories. Existing sharing methods often select or summarize information before other agents' needs become clear, potentially overlooking details that become important later. We propose Ask on Demand (AOD), a communication framework that enables agents to seek targeted peer assistance when specific information needs arise. Upon an approved request, AOD retrieves relevant evidence from peer execution histories, selects a suitable peer, and generates a response tailored to the request. To determine when such assistance is worthwhile, we train a lightweight gate using paired team continuations, with supervision derived from final team outcomes and inference cost. Experiments on AppWorld and WebShop demonstrate that AOD consistently improves task performance over independent parallel execution while achieving lower inference cost than existing information-sharing baselines. These findings highlight the benefits of aligning peer communication with agents' evolving information needs to achieve more effective and efficient collaboration.