Glossary
AI Agency
AI agency is the degree to which an AI system can pursue a goal on its own, planning steps, calling tools and taking actions without a person directing each one; the term is unrelated to marketing agencies that sell AI services.
Term: AI Agency
AI agency is the degree to which an AI system can pursue a goal on its own, planning steps, calling tools and taking actions without a person directing each one; the term is unrelated to marketing agencies that sell AI services.
Why it matters
Agency is a spectrum. Some systems only make suggestions; others decide and act with minimal human intervention. The more a system can do unprompted, the more permissions, monitoring and the ability to stop it matter.
No single settled definition exists. ISO/IEC 22989 defines an AI agent as an automated entity that takes actions to achieve its goals, and keeps ‘autonomy’ for systems that change their own goals. The NCSC describes agentic AI by capability: using tools, making decisions and acting without continuous human intervention.
How it works in practice
An assistant that drafts a reply has low agency; an agent that books meetings and updates records without asking has far more.
As agency rises, human oversight usually moves from approving each action towards human on the loop. That puts more weight on logging and on control of what the agent can reach, such as an MCP server.
Related terms: prompt injection and human agency, the human-side counterpart. Agent security controls sit under AI Security Projects.
Want this in context?
See how this term fits into the broader programme of work.