Summary of "Build & Sell n8n AI Agents (8+ Hour Course, No Code)"

Course Overview

The course is designed for complete beginners with no coding experience.
Goal: Build powerful no-code AI agents and workflows using n8n.
Instructor has earned over half a million dollars teaching AI Agent building.
By course end, students will have 15+ AI automations ready.
The course is comprehensive, timestamped for easy navigation, and covers foundational knowledge through advanced multi-agent architectures.

Large Language Model (LLM) alone: Takes input, generates output (e.g., ChatGPT writing an email), but cannot take actions.
Tools: Integrations like Gmail, CRM, Google Sheets, Airtable, etc. that perform actions.
AI Workflow: Linear, deterministic process combining LLM + tools in fixed steps (e.g., new lead → research → draft email → send email).
- Pros: Reliability, cost efficiency, easier debugging, scalability.
AI Agent: Autonomous, non-deterministic system with a "brain" (LLM + memory) and instructions (system prompt) that decides which tools to use and when.
- More flexible but slower, costlier, and error-prone if used unnecessarily.
Decision: Build workflows first (crawl, walk, run), then move to agents.

Input → AI Agent → Output.
Agent consists of:
- Brain: LLM + memory (short-term and long-term).
- Instructions: System prompt defining role, tools, and usage guidelines.
Memory allows context retention across conversations.
System prompt is static; user input is dynamic.

Sign up for a 14-day free trial.
Dashboard overview: executions (runs), workflows, projects, admin panel.
Workflows consist of triggers + nodes (actions, AI, data transformations).
Credentials (API keys, OAuth tokens) needed to connect to external services.
Nodes have input, configuration, and output panels.
JSON is the core data format; understanding JSON structure and manipulation is crucial.
Binary data represents non-text files (images, PDFs, videos).

RAG: Retrieval Augmented Generation — agent queries a vector database to answer questions.
Vector database (e.g., Pinecone) stores document chunks as vectors based on meaning.
Workflow:
- Google Drive trigger on new file in folder.
- Download file, split into chunks.
- Use OpenAI embeddings to vectorize chunks.
- Upload vectors to Pinecone.
- Chat trigger + AI Agent queries Pinecone to answer user questions.
Tools used: Google Drive, Pinecone, Open Router (for multiple LLM models).

Trigger: New Gmail email.
AI text classifier node categorizes email (customer support, other).
Customer support branch:
- AI Agent uses Pinecone vector DB to find relevant info.
- AI generates email response.
- Gmail node replies to email.
- Optional: Add label to email.
Workflow is deterministic and more cost-effective than agentic approach.

Trigger: New invoice PDF in Google Drive folder.
Download invoice, extract text from PDF.
AI information extractor extracts key fields (invoice number, client name, amount, dates).
Append row to Google Sheets invoice database.
AI crafts email to billing team.
Gmail node sends email.
Workflow handles invoices with varying formats using AI extraction.

APIs = Application Programming Interfaces, enabling communication between systems.
HTTP requests (GET, POST) used to interact with APIs.
Native integrations in n8n wrap API calls with UI; HTTP request nodes allow custom API calls.
Important API call components:
- Method (GET, POST)
- Endpoint URL
- Query parameters
- Header parameters (often for authorization)
- Body parameters (data sent in POST requests)
Use cURL import feature in n8n to quickly set up HTTP requests.
Common API errors:
- 200 = success