I put together a step-by-step walkthrough for building an actual working AI agent in n8n β not just a single AI node, but one that can call tools and reason across steps (Chat Trigger β AI Agent β Tool Workflow).
It covers the part most tutorials skip: a real, complete system prompt you can copy and adapt, plus the most common mistakes (too many tools at once, no human-approval step for anything irreversible, forgetting Memory for multi-turn chats).
Full write-up here: N8n AI Agent Tutorial: Build A ChatGPT Workflow (2026)
Happy to answer questions about the tool-calling setup if anyoneβs stuck on theirs.