Agency is an AI system’s ability to perceive situations, plan toward goals, and act with limited direct human intervention. It shifts AI from producing predictions or replies toward participating in workflows and decisions. section 8.25 section 10.3
Agentic systems can extend assistance, but their autonomy creates practical and social demands:
- They may use context, memory, and reasoning to manage tasks, negotiate responsibilities, and participate in decision-making. section 10.3
- They may coordinate actions across digital or physical environments, including devices, autonomous drones, and smart appliances. section 10.5
- They require safeguards such as sandboxed tool access, behavioral monitoring, user oversight, organizational norms, and clear policy guardrails. section 8.25
- They can create compounding failures, obscure responsibility, and influence outcomes in ways people may not fully understand. section 8.25 section 10.5
From assistant to collaborator
A hybrid human–AI system might jointly plan, execute, and revise a complex task. As delegation becomes more dynamic, people must decide which actions require human approval and how responsibility is assigned. section 10.5
Agency therefore describes more than greater usefulness. It marks a change in how software operates and how humans collaborate with intelligent systems, making trust, authority, alignment, and governance central concerns. section 10.3 section 8.25
