Multi agent prompt design improves coordination and execution separately
CEPO: A General Framework for Improving Prompt Optimization in Multi-Agent Systems
Artificial Intelligence
Summary
Multi-agent systems need both teamwork and special individual skills to get tasks done well. Existing methods mix coordination and task instructions in the same prompts, making improvement hard. The authors introduce CEPO, which splits the team coordination prompts from the individual agent task prompts, allowing better focused updates for each role. Tests on many benchmarks show this split approach improves performance and uses optimization resources more efficiently as teams grow.
What this means in practice
- •For software engineers: Design multi-agent AI systems that coordinate better while optimizing each agent’s task instructions separately for improved overall outcomes.
- •For document processing teams: Improve document understanding systems by optimizing cooperative multi-agent prompts for more efficient and scalable performance.
Authors
Jongha Kim, Seohyun Lee, Jonathan Minho Chung, Jinsung Yoon, Hyunwoo J. Kim
Abstract
Multi-agent systems (MAS) require both effective coordination and specialized task execution. Existing prompt optimization approaches typically couple these functions within agent-specific prompts. We introduce Coordinator-Executor Prompt Optimization (CEPO), a general framework for improving prompt optimization in MAS. CEPO separates system-level coordination from agent-level execution through a shared coordinator prompt and agent-specific executor prompts, and guides their updates according to their respective roles. This structure allows shared coordination rules to be refined consistently across agents while keeping execution instructions separately adjustable. CEPO provides the prompt structure and optimization guidance, while existing prompt optimizers perform the underlying search. Extensive experiments spanning 11 benchmarks across four domains demonstrate that applying CEPO to four existing prompt optimizers consistently improves their performance. Ablations and qualitative analyses support the proposed design, while controlled studies on a document-understanding MAS further demonstrate improved optimization budget efficiency and performance scaling with the number of agents.