Recent AI progress has been measured mostly at the model or application level. Once these AI capabilities enter enterprises, governments, hospitals, laboratories, and other real organizations, however, the surrounding organization becomes the bottleneck. A capable model does not by itself know local knowledge, conventions, authority boundaries, accountability structures, task histories, or heterogeneous systems. Organizational intelligence (OI) is the system-level capability that emerges when AI agents are embedded in, constrained by, and—under governance conditions—able to drive reflection and evolution of a complex organization. It couples people, roles, processes, knowledge, data, systems, permissions, and accountability with agents that perceive, ground, remember, plan, act, verify, collaborate, and learn. The organization, not the isolated model or task, is the right unit for evaluating and governing agentic AI: its most consequential effect may lie less in which tasks are automated than in how intelligence is organized, governed, and improved across an organization. The term has a long genealogy across sociology, management, information systems, collective intelligence, and multi-agent systems; the contribution is a synthesis for the era of large language model agents, not a priority claim. The paper models the organization as a dynamic information-processing and decision network, develops a capability-loop model of nested operational, reflective, and evolution loops, maps the capabilities to implementation components, and traces implications for organizational form and human-AI symbiosis. It then proposes how to evaluate OI, through capability signals and an L0–L5 maturity model whose higher levels demand not just broader autonomy but stronger governance entry conditions: least privilege, audit, independent verification, human control, and controlled evolution.
Xipeng Qiu (Fri,) studied this question.