From Skills to Talent: Organising Heterogeneous Agents as a Real-World Company
Zhengxu Yu, Yu Fu, Zhiyuan He, Yuxuan Huang, Lee Ka Yiu, Meng Fang, Weilin Luo, Jun Wang
2026-04-28
Summary
This paper introduces a new way to build and manage teams of AI agents, moving beyond simple collections of skills to create something more like a self-organizing company.
What's the problem?
Currently, building multi-agent systems is difficult because they're often rigid – the team structure is fixed, agents are tightly linked, and they don't really learn and improve over time as a whole. It's like building a robot team where everyone has a specific job and can't adapt if something changes or a new skill is needed. The core issue is a lack of a system for *organizing* these agents, separate from what each agent individually knows how to do.
What's the solution?
The authors propose a framework called OneManCompany (OMC). Think of it like this: individual agents are 'Talents' with specific skills and tools, and these Talents can be hired and fired from a 'Talent Market' as needed. A central system then orchestrates these Talents through defined roles. Crucially, OMC uses a process called 'Explore-Execute-Review' (E^2R) which is a loop where tasks are broken down, carried out, and then the results are analyzed to improve the process. This ensures the system doesn't get stuck and keeps getting better, similar to how a real company operates.
Why it matters?
This work is important because it allows AI systems to be much more flexible and adaptable. Instead of needing to be pre-programmed for every possible situation, these 'AI organizations' can dynamically reconfigure themselves to tackle new and complex problems. The results show OMC performs significantly better than existing methods, suggesting it's a big step towards creating truly intelligent and versatile AI systems that can handle a wide range of tasks.
Abstract
Individual agent capabilities have advanced rapidly through modular skills and tool integrations, yet multi-agent systems remain constrained by fixed team structures, tightly coupled coordination logic, and session-bound learning. We argue that this reflects a deeper absence: a principled organisational layer that governs how a workforce of agents is assembled, governed, and improved over time, decoupled from what individual agents know. To fill this gap, we introduce OneManCompany (OMC), a framework that elevates multi-agent systems to the organisational level. OMC encapsulates skills, tools, and runtime configurations into portable agent identities called Talents, orchestrated through typed organisational interfaces that abstract over heterogeneous backends. A community-driven Talent Market enables on-demand recruitment, allowing the organisation to close capability gaps and reconfigure itself dynamically during execution. Organisational decision-making is operationalised through an Explore-Execute-Review (E^2R) tree search, which unifies planning, execution, and evaluation in a single hierarchical loop: tasks are decomposed top-down into accountable units and execution outcomes are aggregated bottom-up to drive systematic review and refinement. This loop provides formal guarantees on termination and deadlock freedom while mirroring the feedback mechanisms of human enterprises. Together, these contributions transform multi-agent systems from static, pre-configured pipelines into self-organising and self-improving AI organisations capable of adapting to open-ended tasks across diverse domains. Empirical evaluation on PRDBench shows that OMC achieves an 84.67% success rate, surpassing the state of the art by 15.48 percentage points, with cross-domain case studies further demonstrating its generality.