AI agent
As of:
An AI agent is a system that uses an LLM as its control loop to pursue a goal autonomously across multiple steps: it plans, calls tools, evaluates intermediate results and decides on the next step. This distinguishes an agent from a simple chatbot, which produces exactly one answer per request.
Technically, an agent typically runs as a loop: the model receives the task, the history so far and descriptions of the available tools in its context window, emits structured tool calls — web search, file access or API calls, for instance — and the results are fed back as new input until the goal is reached or a stop condition applies. Standards such as the Model Context Protocol unify how such tools are connected; memory mechanisms and knowledge retrieval via RAG are often added.
In practice, agents are well suited to research, software development, data preparation and process automation. A common misconception: “autonomous” does not mean error-free. Mistakes can compound across many steps, which is why permission models, oversight mechanisms and human approval for critical actions are standard practice. Moreover, not every problem needs an agent — a single, well-crafted model call is often faster, cheaper and more predictable.