What Makes an Agent? Structural Conditions for Artificial Agency

This paper develops a structural account of artificial agency by specifying the necessary architectural conditions under which artificial systems can meaningfully exhibit agency. Building on prior work modeling AI Agent Plasticity as a developmental dynamic (Matta, 2026a), the present study shifts from trajectory analysis to architectural specification. Using a method of systematic structural decomposition informed by ecological psychology, enactivist cognitive science, and dynamical systems theory, the paper identifies five individually necessary and jointly sufficient structural components for the ascription of meaningful artificial agency: (1) a field of affordances constituting the environment of possible action, (2) an embedded system capable of inhabiting that field through perceptual–action coupling, (3) an integrative coordination architecture organizing environmental representation and strategic orientation across temporal scales, (4) a valuation structure generating preference orderings and enabling directed behavior, and (5) a plasticity mechanism enabling structural reorganization over time. The framework introduces a critical distinction between valuation *structure* (architecturally necessary form) and valuation *content* (environmentally shaped specifics). Structural components are derived through an eliminative procedure demonstrating that the removal of any component reduces the system to a recognizable non-agentic category. Case illustrations applying the framework to reinforcement learning agents, large language models, and robotic systems demonstrate its diagnostic application. By distinguishing structural form from environmental content and engaging alternative accounts—including extended cognition, radical externalism, and instrumentalist positions—the framework provides a non-anthropomorphic, structurally rigorous foundation for analyzing artificial agency. The paper establishes an architectural baseline for subsequent dynamic modeling and formal simulation within a developmental systems theory of artificial agency.