Agent
An AI system that takes multiple steps toward a goal — typically by using tools, deciding what to do next, and iterating without human input at each step.
An agent, in current usage, is an AI system that pursues a goal over multiple actions instead of returning a single response. A chat assistant that answers your question and stops is not an agent. A system that reads your question, decides to search the web, decides to write some code, runs it, checks the output, and continues based on what it finds — that's an agent.
The core capabilities that separate agents from chat: tool use (calling APIs, running code, reading files), planning (breaking a goal into sub-steps), and iteration (using the result of one step to inform the next). Recent frontier systems can do all three, though how well they do them depends heavily on the specific task.
The term is used loosely — "agent" often ends up in marketing for products that would previously have been called "workflow automation" or "chatbot with tools". A useful sniff test: does the system make non-trivial decisions between steps? If yes, agent. If it's just following a hardcoded sequence with LLM calls in the middle, it's automation with LLM steps.