A workflow (an automation flow) follows steps someone defined in advance, even if it has branches, loops or a step that uses AI. An AI agent, by contrast, picks the next step based on what it finds along the way. The workflow is more predictable; the agent helps when the steps cannot be planned ahead.
✅ Tests pass
A question from the app. Pick an answer.
In n8n an email comes in, an AI labels it and it goes to sales or support. An agent?
Using AI or having branches is not enough: someone drew every path in advance. n8n can host an agent too, but this is not one.
The AI only picks between two branches already built. An agent would decide what to do next based on what it finds: look up the order, ask or escalate.
A 50 step workflow is still a workflow. What counts is who decides the next step: the design or the system itself.
A workflow: an email comes in, an AI labels it and it goes to sales or support. It uses AI and has branches, but someone drew every path in advance. Tools like n8n, Zapier or Make are used to build flows like this, linking steps on a screen.
An AI agent gets a goal ("resolve this complaint") and decides what to do: it looks up the order, sees it shipped a week ago, checks the delivery and chooses whether to reply, ask something or hand it to a person. Nobody drew that path. It is not about the number of steps: a 50 step workflow is still a workflow.
The workflow, when you know which steps are needed: copying form data into a spreadsheet, posting to a chat when a payment comes in. It does the same thing every time and is easy to test. The agent, when each case needs something different and cannot be planned; in exchange, it is less predictable and you have to check its work. Before adding an agent, compare cost and results with a simple workflow.
Many automations mix both: the complaint comes in, the workflow logs it, an agent investigates in one step, and the workflow asks a person to approve before replying. The predictable parts stay fixed and the agent only works where a decision is needed. The same tool can host both: n8n, for example, has agent steps.
"If it uses AI, it is an agent." No: an AI step inside a fixed path is still a workflow. "An agent is always better." If the steps are known, a simple workflow is usually more predictable. "Built in n8n means workflow; built with AI means agent." What counts is who decides the next step, not the tool.