Abstract
Artificial intelligence is transforming how organizations coordinate, decide, and produce value. Recent advances in large language model-based agent architectures intensify a longstanding shift from centralized managerial control toward distributed intelligence embedded in organizational infrastructures. This paper reframes the “agentic organization” as a networked socio-technical system, rather than a managerial philosophy or automation trend. Drawing on research in multi-agent systems, coordination theory, organizational networks, human-AI collaboration and digital capability, the study develops a conceptual foundation for understanding how work transitions from process-driven execution to agent-mediated decision environments. It presents a three-domain structural model comprising an agentic domain, a human-supervisory domain, and an infrastructural domain, and analytically derives propositions explaining how alignment among these domains, conditions the migration of work toward autonomy, the reliability of agentic workflows and the emergence of systemic risks. A semi-technical scenario illustrates how agentic organizations function as information networks. The analysis demonstrates that autonomy arises not from agents in isolation, but from the interaction of agentic density, supervisory capability, and infrastructural maturity. The paper identifies implications for architecture, capability development and governance, and outlines priorities for empirical research.