Agentic AI
Also known as: AI agent, Agentic artificial intelligence
In short
AI systems that select and carry out steps toward a task, using available tools and feedback within defined permissions.
Agentic AI describes systems that carry out work through a sequence of actions, rather than only generating a single answer. A model can select tools, inspect their results, and adjust the next action within the application’s limits.
The term covers different levels of autonomy. A fixed automation follows predefined steps. An agent has more discretion over which steps to take, although approvals and stopping rules can still constrain it.
A practical example
Imagine a catalog assistant asked to investigate rejected product listings. It retrieves the rejection report, checks the affected records, and compares them with channel requirements.
It then prepares proposed corrections for a specialist to review. If it lacks the missing manufacturer information, it reports that gap instead of inventing a value. This illustrative workflow separates investigation from permission to publish changes.
Autonomy needs boundaries
An agent’s capabilities depend on its tools, data access, and authorization. Access to a product database does not automatically include permission to change prices or submit orders.
Evaluate completed actions and their evidence, including failed tool calls and incomplete work. A fluent explanation of a task is not proof that the task was executed successfully. Human review remains useful wherever an incorrect action would affect customers or money.