Glossary
Agentic AI
Agentic AI describes systems that pursue a goal across multiple steps, choosing their own actions along the way, rather than producing a single response to a single prompt.
What the term is doing
"Agentic" marks a shift in shape, not capability. A generative system answers what it is asked. An agentic system is given an objective and works out the steps, adapting as results come back. The same underlying model can power both; what changes is the loop around it.
What counts
Three properties together: it plans across multiple steps, it takes actions with real effects outside the conversation, and it reacts to what those actions returned. A system missing the third is a script with a model in it.
Why regulators started using the word
Because the governance questions change. A system that acts semi-autonomously raises questions a chatbot does not — who is accountable, what oversight exists, how proportionate the controls are to the risk. Published governance guidance now addresses agentic systems specifically, and the consistent theme is that humans remain accountable for what the system does.
In practice
If someone calls their product agentic, the useful question is what it can do that you would have to undo. That answer tells you whether the label is descriptive or decorative.
Related terms
FAQ
What does agentic AI mean?
Systems that pursue a goal across multiple steps, choosing their own actions along the way, rather than producing one response to one prompt. Three properties together define it: planning across steps, taking actions with effects outside the conversation, and reacting to what those actions returned.
Is agentic AI different from generative AI?
It is a different shape rather than a different technology. The same model can power both. Generative AI answers what it is asked; agentic AI is given an objective and works out the steps. The governance questions differ, which is why published guidance now addresses agentic systems specifically.