Glossary
Also: Agentic workflow
A language model that does not only answer but calls tools over several steps, judges intermediate results and continues.
The difference to a single call is the loop: plan, act, check the result, adjust. That solves tasks needing several steps and external data.
With the loop come new questions: when does it stop, what does a run cost, who is liable for an action, and how do you notice the agent going in circles.
How you notice it
Frequently asked
A cap on steps, a budget per transaction, logging of every tool call, and human confirmation before anything that cannot be undone. Missing one of those makes it a demo, not a system.