AI agent
An AI agent is a large language model running in a loop with access to tools and data until a task is done.
A large language model is the kind of AI model behind ChatGPT and Claude. Tools are the actions an agent can take in other software, such as pulling campaign results or pausing an ad. The agent's own cycle of work is separate from a harness's six loops.
Anthropic defines an agent harness as the system that enables a model to act as an agent. A June 2026 survey of agent research describes an agent as model plus harness. The harness wraps the model, and you can swap the model.
In one engineer's experiments at Anthropic, agents reliably rate their own work too generously on judgment calls such as design, so separate agents run separate checks. An agent can also ignore an instruction, but it can't misuse a permission it doesn't have.
Example
In a paid social example, the drafting agent reads the client's request and each audience's rules. It drafts a brief and an ad, with copy and an image, for each audience. It compares the result with the task and keeps going until the draft is done.