How it works
A chatbot answers one message and stops. An agent works in a loop. It reads the goal, chooses an action such as searching the web or reading a file, looks at the result, and decides what to do next. It repeats this until the goal is met or it gives up. The tools are what make it useful: without them a model can only write text.
The difference from ordinary automation is who chooses the path. In a workflow, you draw every step and the model fills in one of them. In an agent, the model picks the steps. That makes agents flexible on tasks you cannot map in advance, such as researching a topic or fixing a bug, and less predictable, slower and more expensive on tasks you can.
Creators meet agents in coding assistants that edit files and run commands, in research features that browse and summarize, and in the agent steps of automation tools such as n8n. An agent can take a wrong turn and keep going, so the usual safeguards are narrow permissions, a limit on steps or spending, and a person approving anything that cannot be undone.
A worked example
You ask an agent to find five podcasts in your niche that accept guest pitches and to draft a pitch for each.
- The agent searches for podcasts on your topic and opens the results.
- It reads each show's site to check whether guests are accepted, and drops the shows that do not take them.
- It drafts five pitches, each mentioning a recent episode of that show.
- It stops and shows you the drafts. You send them yourself, because sending is the step you did not hand over.
An illustration, not a real account. Any numbers in it are made up to show the method, and are not benchmarks.
Common mistakes
- Using an agent where a fixed workflow would do. If the steps never change, a workflow is cheaper, faster and easier to debug.
- Giving an agent wide access on day one. Start read-only, then add the ability to write or send one permission at a time.
- Treating a confident report as proof. Check the work itself, not the agent's summary of it.
Questions people ask
What is the difference between an AI agent and a chatbot?
A chatbot replies to a message. An agent takes actions with tools, looks at what happened and chooses the next action, so it can finish a task that has several steps.
Are AI agents reliable?
They are improving, but they still make mistakes, and a mistake early in a long task can carry through to the end. Keep a person in the loop for anything that spends money, sends messages or deletes data.
Do I need an agent or a workflow?
If you can write the steps down in order, build a workflow. If the steps depend on what is found along the way, an agent may fit. Anthropic's own guidance is to use the simplest solution that works.
Sources
Anthropic defines workflows as systems where models and tools follow predefined code paths, and agents as systems where models direct their own processes and tool usage. It recommends finding the simplest solution possible.
Anthropic: Building effective agents 19 December 2024
Google defines an agent as software that can reason about user inputs in order to plan and execute actions on behalf of the user.
Every link was opened on 13 October 2026. Products and policies change, so check the source before you rely on a figure.