Fundamentally, "Agent" refers to anything that operates in an "observe-act" loop. So in the context of LLMs, an agent sees an observation (like the code base an

Fundamentally, "Agent" refers to anything that operates in an "observe-act" loop. So in the context of LLMs, an agent sees an observation (like the code base and test output) and produces an action (like a patch), and repeats.