"AI workflow", "agent workflow" and "AI agent" get used as if they meant the same thing. They do not, and the difference decides which tool you should reach for. The short version: the more of the route you decide in advance, the more predictable the result; the more you leave to the AI, the more it can handle things you did not plan for.
AI workflow: you design every step
- 1A new email arrives
- 2AI writes a summary
- 3Summary goes to Slack
Agent workflow: a fixed route, with agents doing the steps
- 1Planner agentsplits the question
- 2Research agentsearches the web
- 3Writer agentdrafts the report
AI agent: it decides the steps itself
- 1You give a goal
- 2Plan, act, checkas many rounds as it needs
- 3Done, or it asks you
An AI workflow: you design the steps
An AI workflow is a fixed sequence that someone set up in advance, with AI doing one or more of the steps. A new email arrives, AI writes a two-line summary, the summary is posted to Slack. Same kind of input, same route, every time.
You build these in automation tools such as n8n, Dify, Langflow or ComfyUI, usually by connecting boxes on a canvas. Because every step is known, a workflow is easy to test, cheap to run, and does not surprise you. The catch is that it only handles what you designed it for; an email in a language you did not expect may come out wrong, and nothing in the workflow will notice.
The Library has ready-made templates you can import, such as Meeting Summary for Langflow, Document Summarizer for Dify and a text-to-image workflow for ComfyUI. Browse all workflows.
An AI agent: it decides the steps
An AI agent gets a goal instead of a route. It works out the steps itself, uses tools such as your files, a terminal or the web, checks each result and tries again when something fails. What is an AI agent? covers this in detail.
Agents handle messy, one-off jobs well: "find out why this script stopped working", "tidy these files and tell me what you changed". They are harder to predict, they cost more per task because they think in rounds, and you want to check their work.
An agent workflow: a fixed route with agents inside
Between the two sits the agent workflow, also called an agentic workflow. The route is still designed in advance, but some steps are handed to agents that can search, reason and retry within their part. Often several agents pass work along: one plans, one researches, one writes.
Langflow's Deep Research Agent template works this way: a planning agent breaks your question into searches, a research agent runs them, and a writing agent turns the findings into a report. You get more judgment than a plain workflow, and more predictability than a single agent left to its own devices.
Which one to use
If the job repeats and always has the same shape, such as summarising every support email or resizing every product photo, build an AI workflow. It is cheaper and easier to trust.
If the job is different every time, use an agent and check what it did.
If the job repeats but one part needs judgment, such as researching a new topic every week and writing it up, an agent workflow is the middle ground: fixed stages, with agents doing the thinking inside them.
Many people end up with all three: a few workflows for the routine, an agent for everything else, and the occasional agent workflow for a recurring report.
Common questions
Is ChatGPT an AI agent? In a browser tab it mostly answers. Newer versions can use tools and act on their own for some tasks, which makes them more agent-like. The line is whether it carries out steps for you or hands you text to act on.
Do I need to code to build an AI workflow? No. Tools like n8n, Dify and Langflow let you connect steps visually, and the Library's templates import in a few clicks. You still need accounts or API keys for the AI services the workflow calls.
Can an agent build a workflow for me? Often, yes. An agent with access to the workflow tool can draft one from a description, which you then test and keep.
Stuck setting one up, or want AI workflows running in your team? Tell us what you are trying to do.