agents capable of planning and executing actions independently across authorized environments. Autonomy in this context refers to an agent’s capacity to decide when and how to act toward a goal, adapting to changing conditions without human guidance. Automation, on the other hand, refers to systems that execute predefined functions reliably under specified conditions without human intervention. The key distinction is that autonomy entails decision- making flexibility (i.e. choosing what to do), whereas automation emphasizes execution reliability (i.e. doing what the system is programmed to do). In the automotive sector, SAE International’s Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles16 framework defines driving automation from Level 0 (no automation) to Level 5 (full automation). A similar spectrum can be applied to AI agents. This spectrum can be conceived of as moving from no autonomy (for example, a simple chatbot that only answers user queries) to full autonomy (for example, a customer service agent that automates interactions, resolves queries and personalizes responses using a company’s knowledge base). Establishing levels of autonomy can help organizations set clear expectations for functionality and implement proportionate governance mechanisms. Authority defines the actions an agent is permitted to take. It sets the boundaries of system access, such as permissions to use tools, interact with databases or execute transactions. Like autonomy, authority exists on a sliding scale, from read-only access to full administrative control. Autonomy and authority can be combined in different ways, depending on an agent’s purpose and design. They are not inherent system properties but design choices that can be made based on the agents’ intended functions, risk considerations and oversight requirements. They can also be calibrated during assessment or adjusted in real time. Operational context refers to the use case and environment in which the agent operates. The environment is especially critical, as it determines observability, predictability of outcomes, interaction with other agents and how conditions evolve over time.17 Use case defines the domain and environment where the agent performs its distinct function for stakeholders. For example, an autonomous cleaning agent in the residential sector performs household vacuuming and floor cleaning as part of routine home maintenance. Environment represents the operating conditions the agent functions under, ranging from simple and predictable settings to complex, uncertain and dynamic contexts. A complex environment is one where the agent navigates and acts under uncertainty, with incomplete or noisy information, unpredictable outcomes, changing conditions over time, continuous ranges of possible actions or states, and interactions with other agents whose behaviour also affects results. By contrast, a simple environment is one where the agent operates with complete information, predictable and static outcomes, independent episodes, a finite set of states or actions, and no need to consider other actors. Establishing levels of autonomy can help organizations set clear expectations for functionality and implement proportionate governance mechanisms. Classification dimensions FIGURE 6 Agent characteristics 1. Function 3. Predictability 4. Autonomy 5. AuthorityWhat does the agent do? 2. Role Deterministic Non-deterministic Low High Low HighOperational context 6. Use case 7. Environment Specialist Generalist Simple ComplexApplication domain and environment where the agent performs its function AI Agents in Action: Foundations for Evaluation and Governance 14

AI Agents in Action Foundations for Evaluation and Governance 2025