A language model on its own can only produce text. Give it access to tools, the ability to run a search, call a service, or perform a calculation, and it can take actions and use the results to decide what to do next. Such a system is often called an agent.
This makes it far more capable, but also riskier, since actions have consequences that text does not. Sensible designs keep powerful actions behind checks, confirm intent before anything irreversible, and treat the model's requests as proposals to be validated rather than commands to obey blindly.