When Should We Orchestrate Multiple Agents?
Multiagent Systems
2025-03-19 v1 Computers and Society
Machine Learning
Abstract
Strategies for orchestrating the interactions between multiple agents, both human and artificial, can wildly overestimate performance and underestimate the cost of orchestration. We design a framework to orchestrate agents under realistic conditions, such as inference costs or availability constraints. We show theoretically that orchestration is only effective if there are performance or cost differentials between agents. We then empirically demonstrate how orchestration between multiple agents can be helpful for selecting agents in a simulated environment, picking a learning strategy in the infamous Rogers' Paradox from social science, and outsourcing tasks to other agents during a question-answer task in a user study.
Cite
@article{arxiv.2503.13577,
title = {When Should We Orchestrate Multiple Agents?},
author = {Umang Bhatt and Sanyam Kapoor and Mihir Upadhyay and Ilia Sucholutsky and Francesco Quinzan and Katherine M. Collins and Adrian Weller and Andrew Gordon Wilson and Muhammad Bilal Zafar},
journal= {arXiv preprint arXiv:2503.13577},
year = {2025}
}