This is a living plan. The next few weeks are fixed. The topics after them are all coming, but in an order we will pick as we go, depending on what the group needs. One rule stays: nothing is introduced before the ideas it depends on.
| Week | Topic | What you will be able to do |
|---|---|---|
| 01 | Introduction to AI | Know what this course is, where AI stands today, and why we chose Claude Code. |
| 02 | Meet Claude Code | Find your way around Claude Code: open it in a folder, use slash commands and keys, watch the harness call tools, switch permission modes, set up a status line. Watch the agent loop run, know which lines to read and how to work safely. See the Claude Code cheat sheet. |
| 03 | Gearing up | Know what a tool is: a button the harness gives the model, and what happens in one tool call. Then give the agent more: a connector (MCP), a command-line program, the service’s API with a small script, or the screen through computer use. Pick the right way for the job: the connector for a quick look, a program or a script for big jobs and files. Know why only Code gets all of them, and what a sandbox is. See the two ways into Gmail in Connect Google Workspace. For homework, do one real task from your job in a project folder: plan first, then Auto. |
| 04 | From skill to repeatable task | Turn a task you do every week into one that repeats. Write a good prompt once (Role, Context, Command, Format), save it as a skill, move the fixed steps into a script, and put it on a schedule. Know why a skill alone is not automation, and which step fits which job. |
| 05 | Becoming the orchestrator | Stop waiting for one agent. Hand side jobs to helpers that each get their own clean context window, run several sessions at once on different tasks, and let Claude run a whole team for a big job. The skills and connectors from weeks 03 and 04 go with every agent. Learn the five habits of the person in charge: split, brief, separate, check, budget. Know when more agents do not help, and why they use up your allowance faster. |
Coming later
In no fixed order.
| Topic | What you will be able to do |
|---|---|
| Context, models and tokens | Understand the context window, why you should never need compaction, and the attractor effect. Pick a model tier from the Claude model family and an effort level, read a benchmark chart, read what a token costs and why you hit usage limits. |
| Files and Markdown | Work confidently with folders and files; open the same folder in Explorer, VS Code and Claude Code. Write notes and documents in Markdown, the plain text format AI reads and writes best. |
| Instructions and checking work | Write instructions that stick (CLAUDE.md) and check the agent’s work. |
| More tools | Image generators, the browser, and other tools the agent can use. |
| Git for humans | Save points and history for your files; share a project with a colleague without emailing zips. |
| Running without you | Let a task run in a loop or on a schedule, even when your computer is off. |
| Working in a team | Shared projects, review each other’s results, keep company knowledge in files. |
| Your own agent setup | The last session. Bring a real task from your job and finish it with the agent, start to end. |
Homework follows each session and builds on the previous one. If you miss a week, read the lecture page and do the homework before the next session; it takes about half an hour.
Related
- Mental models - read before week 02
- What is new - what has actually landed in the wiki so far
- What you will take away - the goals this plan serves
- Lectures - slides and notes for the sessions that already happened