New framework designs flexible yet auditable AI agent orgs

Researcher Lucian Zhu has published a framework for organizing AI agents that mirrors corporate governance structures while enabling real-time adaptability. The paper, titled 'Fluid Structure, Rigid Record: A Layered Organizational Design Framework for Agent-Native Organizations,' proposes a five-layer architecture that separates persistent governance mechanisms from dynamic execution layers.

The framework introduces a four-store record architecture with resident specialization agents, a coordination layer defining agent permissions and privileges, and a runtime layer that dynamically assembles task groups. A human-interaction layer through a control plane provides oversight. Three orthogonal role groups handle operations, review, and supervision with clearly defined scopes and authorities. The design maintains structural rigidity in records and governance while allowing fluid adaptation in team composition and workflows - essentially creating 'operational fluidity with structural permanence.' The prototype has been evaluated in small-scale experiments, though large-scale validation remains pending.