Multi-agent Architecture Search via Agentic Supernet

Large Language Model (LLM)-empowered multi-agent systems extend the cognitive boundaries of individual agents through disciplined collaboration and interaction, while constructing these systems often requires labor-intensive manual designs. Despite the availability of methods to automate the design of agentic workflows, they typically seek to identify a static, complex, one-size-fits-all system, which, however, fails to dynamically allocate inference resources based on the difficulty and domain of each query. To address this challenge, we shift away from the pursuit of a monolithic agentic system, instead optimizing the $\mathbf{\text{agentic}}\mathbf{\text{supernet}}$, a probabilistic and continuous distribution of agentic architectures. We introduce MaAS, an automated framework that samples query-dependent agentic systems from the supernet, delivering high-quality solutions and tailored resource allocation ($\mathit{\text{e.g.}}$, LLM calls, tool calls, token cost). Comprehensive evaluation across six benchmarks demonstrates that MaAS $\mathbf{\text{(I)}}$ requires only $6\sim 45\mathrm{\%}$ of the inference costs of existing handcrafted or automated multi-agent systems, $\mathbf{\text{(II)}}$ surpasses them by $0.54\mathrm{\%}\sim 11.82\mathrm{\%}$, and $\mathbf{\text{(III)}}$ enjoys superior cross-dataset and cross-LLM-backbone transferability.