n8n with AI: automations that think

What you'll see in this lesson

1. n8n: fixed workflows with a brain where needed

   n8n chains apps with visual nodes and lets an AI model decide only in the steps that need judgment.

2. The AI Agent node: model, memory and tools

   The AI Agent node connects a chat model, a memory and some tools, and repeats the think-and-act cycle until it's done.

3. Example: sort emails and prepare the reply

   A typical workflow reads each email, classifies it with AI and sends it down the right branch, leaving the reply as a draft.

4. MCP both ways: client and server

   n8n can use tools from any MCP server and, the other way around, offer your workflows as tools to Claude or ChatGPT.

5. Human approval before sensitive steps

   You can make the workflow or the agent stop and wait for someone to approve on Slack, Telegram or email before acting.

6. Evaluations: check it's right before changing

   Evaluations run a list of test cases through your workflow and score the answers so you can compare prompts or models.

7. Host it yourself with Docker

   n8n's Community edition is free, doesn't limit executions and starts with two Docker commands.

8. What changed in version 2

   n8n 2.0 came out on December 15, 2025, with more security by default and separating saving from publishing.

9. Summary

   n8n automates the fixed parts and leaves only the decisions to AI, with human approval and tests before publishing.