Agentic AI describes systems designed to take actions, not just produce text. Where a plain large language model call turns one input into one output, an agentic system plans, acts, checks the outcome and tries again, often across many steps and several tools.
The word is a spectrum, not a switch. A fixed workflow with one AI step in it is barely agentic. A system that chooses its own next step from a list of tools is much more so. The more freedom the model has, the more useful it can be on messy tasks and the harder it is to predict, test and keep safe.
For builders the practical questions are how much autonomy to grant, what the system may touch, when a human must approve, and how to see what it did afterwards.
Related: AI Agent, Tool Calling, AI Automation, Workflow Automation.