Search papers, labs, and topics across Lattice.
This paper investigates the effectiveness of prompt optimization in multi-agent systems (MAS) composed of LLM-based agents, addressing the complexities introduced by an exponentially growing search space. By systematically analyzing various MAS configurations鈥攊ncluding task types, workflows, communication protocols, and team sizes鈥攖wo advanced prompt optimizers were benchmarked against state-of-the-art single-agent methods. The findings indicate that while prompt optimization can yield substantial performance improvements in MAS, the extent of these gains is highly sensitive to system configuration, highlighting both opportunities and challenges for future research.
Prompt optimization can significantly enhance multi-agent LLM systems, but its effectiveness hinges on the specific configuration of the agents and their interactions.
Multi-agent systems (MAS) offer a scalable path forward for agentic AI, comprising multiple LLM-based agents, each assigned a system prompt and a position within a workflow that governs inter-agent coordination and output aggregation. System prompts thus form a critical and accessible optimization surface: they specify agents' roles and behaviors, enabling system-level improvements without model finetuning. Although prompt optimization has shown substantial potential for single LLMs, extending it to MAS poses distinct challenges, notably an exponentially growing search space. It remains unclear whether, when, and by how much prompt optimization improves MAS performance, and how sensitive such gains are to system configuration. In this work, we systematically study system-prompt optimization across a broad range of MAS setups varying in task, workflow, communication protocol, and team size, benchmarking two prompt optimizers that naturally extend state-of-the-art single-agent methods. The results reveal its potential to unlock significant gains while exposing open challenges, characterizing when and how much prompt optimization helps across diverse MAS settings.
