AI agent: the details
The difference from a classic automated workflow is who decides. In a workflow, the path is written in advance: if an invoice arrives, extract these fields and save them here. In an agent, the model gets a goal and a set of tools—search the ERP, read a PDF, draft an email—and chooses the order. That makes it useful when cases vary too much to anticipate them all.
That freedom has a cost: an agent is less predictable and more expensive to run than a workflow. So it pays to give it few tools, minimal permissions and a step limit, and to write down which actions it may complete on its own and which need confirmation. Many tasks requested as “an agent” are better solved with a workflow and a single model call, or with RAG when what’s missing is context.