Definition
AI that pursues a goal across several steps instead of answering one prompt; choosing what to do next, using tools, and carrying a task through to a finish. In a business setting that means noticing something, gathering the context around it, taking an approved action, and reporting the result, rather than returning text and waiting for a person to act on it.
What it looks like in practice
A landscaping company's system notices that an estimate sent nine days ago has had no reply. Rather than adding a row to a report, it checks what was quoted, confirms the customer has not called since, drafts a check-in in the owner's usual wording, and, because a first follow-up is covered by an approved rule; sends it and logs it. The customer replies asking for a lower price. That is outside the rule, so the thread stops there and goes to the owner as a decision with the history attached. Two steps ran on their own; the third waited for a person.
Why it matters
The word gets used loosely, so it is worth being concrete: the difference is not intelligence, it is whether the system takes the next step or hands you a suggestion. That is also exactly why agentic systems need boundaries. A tool that can act can act wrongly, at speed, in your name, so the useful question about any agentic product is not how capable it is but what it is permitted to do without asking, and what evidence you get afterwards. Marketing copy that avoids that question is describing a risk it has not thought about.