A Functional Taxonomy for Agentic AI: Foundational Characteristics for Evaluating Levels of Agenticness

The emerging paradigm of agentic artificial intelligence challenges current risk management and regulatory approaches. Existing legal requirements and compliance methods primarily rely on risk-based classifications grounded in intended purpose, level of autonomy, and scope of deployment. However, there is no unanimously accepted conceptualization of what constitutes agenticness, including its core characteristics and capabilities. This conceptual gap limits the possibility to comprehensively identify, mitigate, and monitor agentic risks, complicating further compliance with safety and legal requirements. This paper presents a formalized functional taxonomy of agenticness, proposing six non-overlapping, capability-based, and scalable dimensions using fuzzy logic.