What is the difference between an agentic workflow and an AI agent?

Anthropic defines workflows as models and tools orchestrated through predefined code paths, and agents as systems where the model directs its own process and tool use. Microsoft's Agent Framework documentation makes the same split and adds a blunt rule: if you can write a function to handle the task, do that instead of using an AI agent.

CrewAI builds the distinction into its product. Flows are structured, event-driven workflows that manage state and control execution; Crews are teams of autonomous agents. Its documentation recommends starting with a Flow and using a Crew inside a Flow step where autonomy is needed.

Ask these before choosing. The more answers point to the left of each pair, the more the task is a workflow.

Anthropic built its Research feature as agents because research is open-ended: the next search depends on what the last one returned, so the path cannot be hard-coded. By contrast, a weekly report that always gathers the same figures, summarises them and sends them to the same people is a workflow — the model writes the summary, but code decides everything else.

Workflows are cheaper, faster and easier to test, because every run follows a known path. Their weakness is rigidity: when an input does not fit the path, the workflow fails or produces nonsense.

Agents adapt, but each extra loop is another model call, another chance for error, and more tokens. Anthropic reports that agents typically use about four times the tokens of a chat interaction. Many good systems are workflows with one agentic step inside them.