Paper Claims Six Topologies Suffice for LLM Multi-Agent Systems

Elvis Saravia posted about the arXiv paper Codebook Agent: Amortized Topology Design for LLM Multi-Agent Systems by Jinxi Yu, Eric Hanchen Jiang, Kai-Wei Chang, Ying Nian Wu and colleagues at UCLA. The work states that adapting communication topology per query improves accuracy and efficiency in multi-agent LLM systems. Their approach grows codebook capacity from 8 to 64, after which topologies surviving a reward filter collapse to roughly six distinct ones. The paper is linked on arXiv and dair.ai.