Course · Beginner to intermediate · 28 lessons · ~5 hours

Claude for Real Work

Part one is Claude the app, and needs no terminal. Part two is Claude Code, on a real codebase with real mess in it - tests that fail, a file that is too long, and a README that lies.

Start lesson 1

What you'll learn

  • What a language model can and cannot see, and why long conversations go wrong
  • Getting a useful answer on the first ask instead of the fourth
  • Running Claude Code on a codebase you did not write
  • Project memory, context and cost: the three dials that decide what a session costs
  • Skills, commands, hooks and MCP, and when each one is the wrong reach
  • Catching the mistakes it makes confidently

Before you start

  • Part one: nothing. If you can use a browser you can do it.
  • Part two: a terminal, git, and a language you can read

Beginner to intermediate · 28 lessons · ~5 hours · Free · No sign-up

  1. Part 1 Claude, the app No terminal, no code. What the thing actually is, how to ask it for work, and where it lets you down.
    01
    9 min

    What Claude actually is

    A model, wrapped in an app that decides what it can see. Getting that one distinction right explains most of what follows.
  2. 02
    8 min

    Getting set up

    The web app, the desktop app and the phone, what each is good for, the two settings worth changing on day one, and what incognito is not.
  3. 03
    12 min

    Asking for things properly

    The four things a request needs before it can go right: the goal, the constraints, an example, and what finished looks like.
  4. 04
    10 min

    What Claude can see

    The context window, in plain terms. Why a long conversation gets worse rather than better, what the tells are, and what to do about it.
  5. 05
    11 min

    Projects

    Instructions and reference material that stay put, so you stop re-explaining your job at the start of every conversation.
  6. 06
    10 min

    Files and documents

    Uploading PDFs, spreadsheets and images. What it reads well, what it reads badly, and how to tell which one just happened.
  7. 07
    9 min

    Artifacts

    When Claude builds the thing instead of describing it, how to steer something you can see on screen, and what publishing one really means.
  8. 08
    11 min

    Connectors

    Wiring Claude into the tools you already use, what access that grants it, where the data lands afterwards, and the ones worth thinking twice about.
  9. 09
    12 min

    Where it goes wrong

    Where Claude goes wrong now, which is not where the folklore says. Three failures that no longer reproduce, and the completeness failure that does.
  10. 10
    9 min

    Making it a habit

    The jobs Claude is good at, the ones it is not, how to keep score for a fortnight, and how to tell from your own use rather than mine.
  11. Part 2 Claude Code For developers. Every exercise runs against stockroom, a small public repo built for this course, and every check is something you can see on your own screen.
    11
    11 min

    Installing, and your first session

    Getting Claude Code installed and running, cloning the repository this half of the course lives on, and asking it something you can check yourself.
  12. 12
    13 min

    The change loop

    Ask, read the diff, accept or push back. The loop you will spend every session in, practised on a bug with a failing test.
  13. 13
    12 min

    Reading before writing

    Using Claude Code to read code you did not write, and checking its explanation against the source rather than believing it because it sounded sure.
  14. 14
    11 min

    Plan mode

    When stopping to plan pays for itself and when it is ceremony. The same task run both ways, with the diffs side by side.
  15. 15
    13 min

    CLAUDE.md

    Project memory for Claude Code: what belongs in CLAUDE.md, what does not, where the file lives, and how to tell whether yours is doing anything.
  16. 16
    12 min

    Context is the bill

    How a Claude Code session fills its context window, what that costs you in money and in quality, and the two commands that give the room back.
  17. 17
    12 min

    Models and effort

    Two dials, one task, and real numbers. What the expensive model and effort setting buy you, and the specific way the cheap one fails.
  18. 18
    12 min

    Permissions

    Deciding once what Claude Code may do without asking, where the permission rules live, and the short list of things that should always stop and ask.
  19. 19
    12 min

    Git

    Branches, commits and pull requests in Claude Code, with Claude doing the typing and you doing the deciding: what to delegate and what to keep.
  20. 20
    11 min

    Your own slash commands

    The thing you type out three times a week, written down once as a slash command. Arguments, injected command output, and where the files live.
  21. 21
    14 min

    Skills

    Skills in Claude Code: instructions that load only when they are needed, how to write a description that triggers, and how reliably that happens.
  22. 22
    12 min

    Subagents

    Handing work to a second Claude with its own context. What you get, what it costs, and the shape of task where a subagent makes things worse.
  23. 23
    12 min

    Hooks

    Making the harness do something every time, whether the model cooperates or not. A rule that redirects rather than just refusing.
  24. 24
    13 min

    MCP

    What the protocol is under the branding, connecting a server to Claude Code, and the question worth asking before you connect one at all.
  25. 25
    11 min

    Plugins

    Packaging the command, the skill, the agent and the hook you just wrote into one thing your team can install, and checking it works for them too.
  26. 26
    12 min

    Headless and CI

    Running Claude Code with nobody watching: one-shot invocations, output a script can parse, a mode that cannot hang, and a review that fails a build.
  27. 27
    13 min

    Catching its mistakes

    The failures that get through review because they look right. A bug with no failing test, and the one habit that finds it.
  28. 28
    14 min

    A week in the life

    One real feature on stockroom, start to finish, using everything from part two. Then the four habits to keep and the five to stop doing on Monday.

Common questions

Is Claude for Real Work free?

Yes. All twenty-eight lessons are free, with no sign-up. Part two uses Claude Code, which needs a paid Claude plan - the course itself costs nothing.

Do I need to be a programmer?

Not for part one, which covers Claude the app and needs no terminal. Part two is Claude Code on a real codebase and assumes you can run a command and read a diff.

How long does the course take?

About five hours of reading across twenty-eight lessons, and every lesson ends with something you do on your own machine rather than something you watch.

What codebase does part two use?

A small inventory tool called stockroom, cloned from GitHub in lesson 11, with real bugs planted in it so the exercises look like the work you actually get paid for.
Start lesson 1 ← All courses