Reproduce “From Skills to Talent: Organising Heterogeneous Agents as a Real-World Company” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zhengxu Yu, Yu Fu, Zhiyuan He, Yuxuan Huang, Lee Ka Yiu, Meng Fang, Weilin Luo, Jun Wang

OneManCompany (OMC) is an open-source framework that models multi-agent systems as self-organizing and self-evolving AI organizations, enabling them to tackle open-ended, real-world projects. The framework achieved an 84.67% success rate on the PRDBench software development benchmark, surpassing state-of-the-art baselines by 15.48 percentage points.

Key insights

  • Decoupling agent capabilities from organizational management through a Talent–Container architecture enables dynamic workforce assembly, reusability, and seamless interoperability of heterogeneous AI agents.
  • Formalizing project execution as an E2R tree search with explicit review gates significantly enhances quality control and task completion reliability, particularly for complex, long-horizon projects.
  • Implementing continuous self-evolution at both individual and organizational levels allows AI workforces to adapt, learn from experience, and maintain performance over time without constant human intervention.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.