What does Agent mean?

An agent is an AI system that pursues a given goal independently across multiple steps, using various tools along the way to reach that goal. The agent plans the next step, carries it out, evaluates the result, and then decides whether to continue or stop. Unlike a chatbot, it does not answer a single question but works a task through to a result.

An agent is built on a language model that accesses tools through a defined interface, for example a line-of-business system, a document archive, or a search function. The model receives the list of available tools along with a description of their parameters and chooses among them itself. After each call, the result flows back into the context, and the loop begins again. A stopping condition limits the number of steps, the runtime, and the cost.

An agent works well wherever a task consists of many small checks and the relevant information sits in separate systems. Typical cases include checking incoming applications for completeness, researching contracts and rulings, and pulling together input from several departments. The agent handles the steps up to a draft, not the decision itself.

The advantage over a plain search function lies in the outcome. An agent does not return a list of matching documents but a finished draft. Work then starts at the decision rather than at the research. A logged process also makes it possible to verify afterward which source fed into which result.

An agent should use only the tools a given task requires, and every process needs a defined stopping condition. In regulated fields such as public administration, healthcare, or finance, there is an additional requirement: the model, its context, and its logs must not leave the applicable legal jurisdiction.