In short
- An agent is a model that plans steps and uses tools by itself instead of only answering a question. In a loop it reads tool results and decides what to do next.
According to Anthropic’s write-up on effective agents, an agent is a setup in which the model dynamically directs its own process and tool usage, keeping control over how it accomplishes the task. It differs from a workflow, in which the model and tools are led along a path written in code. The agent’s working pattern derives from research on interleaving reasoning and acting, described in the ReAct paper.
When an agent pays off
An agent is better when flexibility and model-driven decisions are needed, and a workflow when the task is well defined. Agents cost more time and money, so start with the simplest solution, as in the article How a coding agent works: the tool loop.
Where to read next
The concepts MCP and workflow. The whole topic: the Coding agents hub.