AI agent — a system in which a language model decides its own next steps, and which tools to use, to reach a goal — rather than following a path someone wrote in advance.
The line between an agent and an automation
The most useful definition comes from Anthropic’s engineering team, which separates workflows — “systems where LLMs and tools are orchestrated through predefined code paths” — from agents, in which the model directs its own process and tool use.
That distinction is worth more than any marketing label. A system that follows the same five steps every time is a workflow, even if one of those steps uses a language model. A system that decides at runtime whether to search, ask, draft or stop is an agent.
When an agent is the right call
When the task is open-ended and the steps cannot be written down ahead of time: research, triage of unusual requests, working across several tools to answer one question. When the steps can be written down, a workflow is cheaper, faster and far easier to audit. The same guidance recommends starting with the simplest thing that works and adding complexity only when it earns its place.
Agents take actions, so their failures are actions too. Anything an agent can do on your behalf — send, book, refund, delete — needs a permission boundary and a person who reviews it. See human-in-the-loop and prompt injection.
Source: Anthropic, Building effective agents, December 2024.
Related: workflow automation, human-in-the-loop, Model Context Protocol, agentic commerce. Longer read: Automation or AI agent?