The Complete Claude Curriculum in 3 minutes

Why it matters: if your team is still typing questions into a chat box, you are using about a tenth of what you pay for. This episode is the map of the other nine tenths, and the one to send someone before they start.

The trust ladder: chat, Cowork, Claude Code, MCP

There are four ways to work with Claude and they are not interchangeable. Chat answers you. Cowork does the work inside the files and tools you already use. Claude Code builds and changes software. MCP connects any of them to the systems your business runs on.

The order matters because each step hands over more. You climb the ladder as your trust in the output grows, not because a rung is "more advanced".

What an agent is, and the loop inside it

An agent is a model in a loop: look at the goal, pick a tool, read the result, decide again. Everything that feels like magic in a demo is that loop running fast. Everything that goes wrong is the loop running without a stop condition.

Workflow vs agent, and the hybrid most businesses use

A workflow is predictable and cheap. An agent is flexible and expensive. Almost every system we ship is a workflow with one agent step in the middle, where judgement is actually needed, and plain code either side.

MCP: the USB-C port for AI

One standard connector between a model and a tool. Build the server once and every Claude surface can use it. This is the piece that turns "Claude knows things" into "Claude can do things in our systems".

Permissions: three incidents, three guardrails

Three real incidents, each stopped by a setting that takes a minute to apply. The video names them; the guides below walk through where each one lives.

The model ladder and what it costs

Haiku, Sonnet and Opus are not good, better, best. They are three price points for three kinds of work, and most tasks belong on the cheapest rung that gets them right. The curriculum guide has the current prices, checked October 2026.