From Skills to Talent: Organising Heterogeneous Agents as a Company [pdf] (arxiv.org)

A recent paper titled "From Skills to Talent: Organising Heterogeneous Agents as a Real-World Company" introduces a novel framework known as OneManCompany (OMC) designed to enhance the functioning of multi-agent systems (MAS). Traditional MAS often face limitations due to fixed team structures and session-bound learning processes. OMC aims to address these constraints by introducing a dynamic organisational model for agents—termed "Talents"—that allows for agile team configurations and on-demand recruitment via a community-driven Talent Market. This innovation signifies a major shift toward more adaptable and capable AI systems that can intelligently reorganize in real-time to meet varied task demands. The OMC framework employs an Explore-Execute-Review (E²R) tree search mechanism, unifying the planning and execution stages while ensuring systematic improvement through iterative feedback loops akin to human management processes. Remarkably, OMC has demonstrated an impressive 84.67% success rate in empirical tests, surpassing current state-of-the-art systems by 15.48 percentage points. This advancement not only illustrates OMC's practical applicability across diverse domains but also highlights its potential to transform static AI systems into agile, self-organizing entities, paving the way for more sophisticated applications in AI and machine learning.