A system that runs a model in a loop: planning, calling tools, and reacting to results to pursue a goal.
Broad technical consensus on what this term means. Disagreements that invoke it are about the world, not the word.
An agent wraps a model in a loop: given a goal, it plans, takes an action (usually a tool call), observes the result, and decides what to do next, across many steps without a human between them. That loop is the difference between answering a question and completing a task.
The catch is compounding reliability: a step that succeeds 99% of the time fails alarmingly often across a fifty-step chain. Much of the labor-displacement and valuation debate is really a debate about whether agents reach unsupervised reliability in high-stakes work, which is why "agents" appear all over the Argument Map's tests.