An AI agent is software that uses a language model to pursue a goal by deciding its own steps: it plans, calls tools (search a system, read a file, draft a message, update a record), observes what came back, and adjusts until the goal is met or it needs help. The distinguishing trait is the loop. Given "compile the weekly report", an agent decides what to read, in what order, and what to do about a gap - where ordinary software follows a path someone scripted in advance.
Under the hood, most agents are the same shape: a language model, a set of tools it's allowed to call, and a loop that feeds each result back into the next decision, bounded by guardrails - which tools, whose permissions, what needs sign-off. Everything else (memory, planning styles, multi-agent teams) is elaboration on that base. So when a vendor says "agent", the useful questions are concrete: what tools can it call, under whose permissions, and what happens before an action leaves the building.
