This research introduces an automated approach to design multi-agent systems (MAS) powered by large language models, optimizing both prompt engineering and agent interaction patterns.

Key innovations:

• Systematic exploration of the multi-agent design space, identifying critical factors affecting performance
• Novel optimization framework that automates both prompt creation and interaction topology design
• Demonstrated performance improvements across complex collaborative tasks
• Engineering-focused approach that reduces manual design effort while improving system efficiency

Business value: This research enables more efficient development of complex AI systems that can tackle collaborative problems, reducing engineering overhead while improving performance and reliability in production environments.

Multi-Agent Design: Optimizing Agents with Better Prompts and Topologies