What is an AI Agent? — Claude & Co-work

Taught by Raul

Chatbots only answer questions. Agents take actions — read Gmail, use the browser, organise files, run pipelines. Raul shows Claude Co-work live: LinkedIn job applications, phone → Downloads cleanup, and a full content workflow (including dispatch, skills, scheduling).

• LLM vs agent — question-answering vs doing things for you
• Skills (.md), dispatch (parent + parallel sub-tasks), scheduled runs
• Koda stack + media pipeline hooks (e.g. Remotion / Fal-style automation ideas)

This is where your mind gets blown 🤯

Takeaway: agents turn AI from a chat window into a personal assistant that can act while you’re away.

Docker + N8N — Build Your First Agent

Mystery co-instructor · identity revealed Session 5

You install Docker Desktop and run N8N locally (free image — no card). Then you watch three real automations, then wire your own first flow in the visual editor.

• Demos: long YouTube → infographic; Gmail assistant (read / send / calendar); inbox sort + auto-reply
• Triggers, nodes, 700+ app integrations — LLM + memory + tools as the agent shape
• Build: trigger → AI node → Gmail (or similar) action

Takeaway: N8N in Docker is a free, private automation engine — set it up once, keep building.

APIs & Webhooks — Connecting the World

Raul (APIs & MCP) + mystery instructor (webhooks & connectors)

APIs are how two apps talk (like Uber pinging Maps). MCP is the AI-native layer so models use pre-built tools instead of reinventing glue code. Webhooks are the doorbell: your workflow wakes up when data arrives.

• GET vs POST in plain English (read the menu vs place the order)
• Webhook test URL vs production in N8N
• Live shortcuts: Claude connectors → Canva poster + Spotify playlist

Takeaway: APIs let your agent talk out; webhooks let the world talk back in real time.

Supabase — Memory, DB & RAG

Taught by Raul

Real products store data. Supabase (Postgres under the hood) is your agent’s memory: signups, chat history, documents. You’ll connect a form, use the SQL editor, and plug vectors + embeddings so the agent can answer from your files (RAG) via N8N.

• Tables, auth, RLS basics, vector search
• Upload a doc → query it through the agent
• Optional: run Supabase locally with Docker for offline practice

Takeaway: Supabase gives your agent durable memory — users, docs, and smart retrieval.

Google AI Studio — Landing Page Build-Off

Mystery co-instructor + Raul (friendly build-off)

Google AI Studio turns plain-English descriptions into working UI (Gemini). You’ll see prompt chaining — use Claude to write the perfect brief for Gemini — plus 21st.dev prompts for hero, features, testimonials, nav, and polish (Pinterest / Spline-style inspiration).

• Land the key sections: hero, features, social proof, FAQ, footer
• Vibe coding: describe it → ship it — no manual HTML/CSS required

Takeaway: you can ship a credible, animated landing page in under an hour with AI.

Cursor — Chat Front-End for Your Agent

Taught by Raul

Cursor is an AI code editor: you describe the UI, it edits the files. Raul hooks your N8N public webhook into a React chat page, routes questions through Supabase RAG, and drops in premium sections via 21st.dev prompts — still without hand-writing code line by line.

• Local dev server: see N8N ↔ Supabase ↔ UI working together
• Next step preview: deploy in Session 7

Takeaway: Cursor is the bridge from backend agent to shareable chat UI.

Deploy Free — GitHub + Vercel

Mystery co-instructor · identity revealed Session 5

If it only runs on localhost, no one else can use it. This session walks through the fastest free path: push from Google AI Studio → GitHub, then import to Vercel → Deploy for a public URL in about a minute. Plus custom domains so it looks client-ready.

• Why hosting matters for sharing & selling a build
• Course docs include a troubleshooting walkthrough

Takeaway: GitHub + Vercel ≈ free, pro deployment in minutes — no server rent.

Prompt Engineering — Quality on Demand

Taught by Raul

Everything you’ve built rides on how you ask. Raul teaches RTOT — Role, Task, Output, Tone — plus three practical modes: zero-shot (no examples), one-shot (one example), and chain-of-thought (step-by-step back-and-forth). Expect live examples + quick challenges.

• Role = who the model acts as · Task = what to do · Output = format · Tone = voice
• Few-shot pattern for trickier requests

Takeaway: better prompts → better outputs — RTOT is your default checklist.