Concept / Term · New (AI-era)
AI that takes actions
An AI system that does not just answer, but carries out multi-step tasks using tools, such as reading files, calling APIs or sending messages. This is why approval controls on what an agent may do matter.
Field note
An AI agent uses a model to choose and sequence actions toward a goal, often calling tools, inspecting results and revising its next step. “Agentic” describes systems with more of this autonomous, multi-step behaviour.
Software agents are older than generative AI, but the language surged as large language models became able to plan with natural-language context and operate general-purpose tools.
Useful agents work within explicit permissions, expose their actions, stop for approval at important boundaries and recover safely when a tool or assumption fails.
The value comes from completing workflows rather than only producing text; the risk comes from converting model mistakes into real actions.
In a sentence“The agent drafts the release note and opens a review, but a person approves publication.”