noun · AI concept
Agent
also called AI agent
An AI that works in a loop. It looks, decides, does something, checks the result and goes again until the job is done.
A model wrapped in a loop with access to tools, which it uses over several steps to work toward a goal instead of answering once.
A harness repeatedly calls the model with the task, the history and the tool results so far. The model either asks for another tool call or gives a final answer, and the loop ends when it answers or a limit is hit.
Real-life example
You ask an assistant to find why a test is failing. It reads the test file, runs the test, reads the error, opens the file named in it, edits a line and runs the test again. You typed one sentence. It made a dozen separate model calls, each one seeing what the last step turned up.
Where the analogy breaks
-
An agent is not one smart thing working on its own. It is an ordinary model called many times by a loop that feeds each result back in.
-
More steps do not mean better answers. Each step can add a small mistake, and a wrong early step can send the rest of the run the wrong way.
-
It costs more than a single question. Every step resends the growing history, so long runs use a lot of tokens and can hit the context window.