Setting Up Claude Code

Full free course here: https://ai.danielpoon.workers.dev/skool

Outcome: Claude Code Desktop installed, terminal comfortable, VS Code or Cursor set up, and your first real conversation with Claude logged inside your own workspace.

Why this module exists

Eighty-four percent of the population still isn't using AI properly. Of the 16% who are, most are stuck in the browser, pasting context into a fresh chat every single session.

The biggest blocker isn't intelligence. It's the terminal.

People see the terminal, their heart rate spikes, and they retreat to claude.ai. They stay there forever — fiddling with prompts, copying and pasting context, wondering why AI isn't changing their life.

We kill that blocker in week one, once you're familiar with terminal, I'll explain why my preferred way of interacting with Claude is now Claude Code Desktop App and occasionally VScode.

Claude.ai in the browser:

• Can't read your folders
• Can't edit files on your machine
• Can't run terminal commands or scripts
• Can't schedule jobs or run in the background
• Loses context the moment you close the tab

Claude Code Desktop:

• Reads any folder you point it at
• Edits files in place
• Runs commands, scripts, and scheduled jobs
• Holds persistent memory through CLAUDE.md and skills
• Dispatches subagents and parallel workers
• Connects to the rest of your life through MCPs (calendar, email, Notion, Slack, whatever you use)

Every lesson from here on assumes you're in Claude Code Desktop. Every workspace.

"But I'm not a developer"....Read This Before You Panic

You don't need to be. I built this course for business owners, creators, and solo operators. You will not write code in this course. You will type commands, the same way you type URLs into a browser.

If you can use a browser, you can use Claude Code. The terminal is just a different-looking browser: text in, text out. Lesson 1.2 makes you fluent in the eight commands you'll ever need. That's it.

When you use Claude in the browser at claude.ai, you're using a chat app. Beautiful UI. Clean conversation. Works on your phone etc. It feels like chatting with ChatGPT or Gemini, except maybe just a little better and less cringey emojis.

However to get the most of AI Large Language Models (LLMs), you need to use the LLMs within a harness that will be able to use the LLMs (think of it as a brain) and to give it tools (think of your AI's having hands) to interact with apps, browsers and your PC by either typing or clicking around and seeing your screen.

The most popular ways are with Claude Cowork and Claude Code.

Above is the top left panel of the CLAUDE DESKTOP APP. This is the chat function (question and answers to Claude. People use this to chat to Claude as therapists, advisors, search engines etc. the most basic way of talking to the Claude "Brain")

Above is Claude Cowork. This is the least intimidating way of chatting to the Claude Brain and getting it to do things in your PC, apps etc. It is all done through the Claude Desktop App. However, I will not be really covering this as I feel it is less flexible than Claude Code and learning Claude Code offers much more customising, flexibility, and future proofing our automations in the future :)

Above is Claude Code. Simply click on it or use the hotkey (Cmd + 3) on Mac. All our lessons are done either here or via terminal (more on this later)

Step 1 — Install Claude Code Desktop

Unfortunately I will only be showing you how to do this via Mac. I believe Mac is the best OS for running Claude Code. They usually ship updates first for Mac and Mac is much more common as a productivity tool especially for coding with software businesses and developers.

Then sign up for Claude Subscription. $20USD if paid monthly $17USD if paid yearly.

Download Claude Desktop via the link here: https://claude.com/download

Step 2 — Once Downloaded,

First launch, Claude Code walks you through setup. Let it run and follow instructions.

For most of you, you'll click Personal Use.

After this you'll be prompted to pay for subscription if you havent done so already. Unfortunately there aren't any free trials for claude code. You may see there's a free tier. but free tier doesn't include claude code or cowork unfortunately :(

Go Deeper

• Official Claude Code docs: docs.claude.com/claude-code
• If you want the full argument for why the shift from chat app to runtime is the biggest AI unlock of 2026, hold the question for Lesson 3.2 — The Folder Is The App. The whole case is laid out there.

1.2 — The Three Tabs

Outcome: A clear mental model for when to use each of the three tabs, so you never waste a session in the wrong one.

The Lesson

Open Claude Code Desktop and look at the top. You'll see three tabs: Chat, Cowork, Code. They look similar. They are not.

Chat — the one that feels familiar

Chat is claude.ai inside the app. General conversation, no file access. Use it when you just want to ask a question and get an answer — the same way you'd use ChatGPT.

Fine for:

• "Explain this concept to me"
• "Draft me an email in the abstract"
• "What's the difference between X and Y?"

It can't touch your files. It can't run anything. It's a chat app with a new coat of paint.

Cowork — the cloud autonomous agent

Cowork is an autonomous agent that works in the cloud on its own VM. You hand it a task and it goes away for twenty minutes and comes back with the result. It runs even if you close the app.

Great for:

• "Research five competitors and summarise their pricing"
• "Read these three PDFs and pull out the common themes"
• "Draft a long report I can review in the morning"

The killer feature: you can shut your laptop and walk out the door. Cowork keeps going. Cowork is your remote worker.

Code — the one that matters most for this course

Code is an interactive coding assistant with direct access to your local folders. You pick a folder. You type a task. Claude reads your files, proposes changes, and you review every edit in the diff view before it touches anything.

This is where you'll build your system:

• Your CLAUDE.md files
• Your folder architecture
• Your skills
• Your content drafts
• Your briefs
• Your scheduled tasks

80% of this course lives in the Code tab. Get comfortable here and the rest of the system falls into place.

Rule of thumb — which tab do I pick?

• Quick question, no files involved → Chat
• Research task that takes 20+ minutes, you want to close your laptop → Cowork
• Building, editing, automating anything on your machine → Code

If you're ever unsure, default to Code. Most real work — drafts, edits, setup, configuration — lives there.

Why this matters

Most people who try Claude Code Desktop and "don't get it" are in the wrong tab. They open Chat, ask it to edit a file, it can't, they conclude the app is broken and go back to claude.ai.

The tab is the difference between a chat app and a runtime. Pick the right one and you're operating. Pick the wrong one and you're back to typing into a box.

Exercise

Click each tab once. Notice how the interface changes:

• Chat — a single clean conversation window. No folder picker. No file access.
• Cowork — a task inbox. You hand off work and check back later.
• Code — a folder picker, a model dropdown, a permission mode selector, and a prompt box. This is your cockpit.

Close Chat and Cowork for now. Pin your attention to the Code tab. That's home base for the next nine modules.

1.3 — Your First Code Session

Outcome: You pick a folder, pick a model, type your first task, and watch Claude propose real work on your real files.

The Lesson

Open Claude Code Desktop. Click the Code tab. You'll see a prompt box with four things to configure before you hit Enter.

1. Environment — where Claude runs

You'll see three options:

• Local — runs on your machine with direct access to your files. Pick this.
• Remote — runs on Anthropic's cloud. Sessions continue even if you close the app. Covered in Module 10.
• SSH — connects to a remote machine over SSH. For devs with their own servers. Covered in Module 10.

For everything in Modules 1 through 9, you'll be on Local. Don't overthink this.

2. Folder — where Claude works

Click Select folder and pick a real project you care about. Not a test folder. Not an empty one. A real project.

Why real? Because you need to feel what this tool can do. Picking a dummy folder is like test-driving a Ferrari in a parking lot — you'll never hit the point where your jaw drops.

Good first folders:

• A blog or website you maintain
• A notes folder with markdown files
• A side project you've been neglecting
• A freelance deliverable you're working on

Windows note: Git must be installed for Code tab sessions to work. If Claude complains, download Git for Windows, install it, restart the app. Most Macs have Git already.

3. Model — which Claude

Click the model dropdown next to the send button.

• Sonnet — fast, capable, cheap. Start here. Use it for 90% of work.
• Opus — slower, smarter, more expensive. Swap in when you hit something complex (big refactor, strategic planning, hard debugging).
• Haiku — tiny, fast, cheap. Good for high-volume scraping and classification jobs. Not for real work.

You can change the model mid-session. Don't obsess over this now.

4. Permission mode — how much autonomy Claude has

Leave this on Ask permissions for your first few sessions. It's the default. It means Claude shows you every proposed change in a diff view before touching your files. You approve each change.

You'll learn the other modes — Auto-Accept, Plan, Auto, Bypass — in Module 2. For now: Ask permissions.

Type your first task

Keep it small. The point is to see the whole loop in under five minutes. Good first tasks:

• "Find a TODO comment in this project and fix it."
• "Add a short README section explaining what this folder does."
• "Rename all .txt files in this folder to .md."
• "Make a list of every markdown file in this folder and save it as index.md."
• "Fix any typos you find in the README."

Press Enter.

What happens next

Claude will:

1. Read your files. You'll see it open a few. That's it gathering context. Don't panic. It's not sending your files anywhere beyond the Anthropic API.
2. Think. You'll see a summary of what it plans to do.
3. Propose changes. A diff view appears — red lines for removals, green for additions. Nothing on disk has changed yet.
4. Wait for you. Accept or reject each change.

That's the loop. Read → think → propose → wait. You're not racing the AI. You're directing it.

Common first-session mistakes

1. Folder too big. If you point Claude at a massive repo with 10,000 files, it'll burn tokens reading context and you'll get slow first responses. Start with a focused folder — a few dozen files, not a few thousand.
2. Task too vague. "Make this better" will give you mushy output. "Fix the typo on line 12 of README.md" or "Rewrite the intro paragraph to be shorter" gives you crisp output. Specificity compounds.
3. Panic when Claude reads files. This is normal. This is the whole point. The tool is supposed to read your files — that's how it gives you real answers instead of generic ones.
4. Hitting Opus for everything. Opus is great for hard problems. Overkill for typo fixes. Start on Sonnet.

Exercise

1. Open the Code tab.
2. Pick a real folder you care about.
3. Leave Environment on Local. Model on Sonnet. Permission mode on Ask permissions.
4. Type one of the prompts above.
5. Hit Enter. Watch Claude read, think, propose.
6. Do not accept or reject anything yet — that's Lesson 1.4.

When the diff appears, pause. Look at it. Notice that nothing has changed on disk. Notice that you're in full control. Feel that.

Outcome: You can read and write markdown well enough to produce any file in this course — CLAUDE.md, context.md, notes, briefs, the lot. Five minutes. Six symbols. Done.

Why markdown?

Every file in this course is a markdown file. Every skill. Every brief. Every context file. Your CLAUDE.md. Your voice.md. Your samples.md.

Why? Because markdown is:

• Plain text — no app needed to open it, no corruption, readable in any editor
• Claude-native — Claude reads, writes, and reasons about markdown better than any other format
• Git-friendly — diffs cleanly, merges cleanly, version-controls cleanly
• Future-proof — if the tool dies, your files don't

Docs go stale. Notion apps shut down. Markdown lives forever. Write everything in markdown.

The six symbols that cover 95% of what you'll ever need

Open any text editor. Type these exactly. See what comes out.

# Heading 1
## Heading 2
### Heading 3

A normal paragraph. Write however you want.

**bold text** and *italic text*.

`inline code` for short snippets.

- Bullet point one
- Bullet point two
- Bullet point three

1. Numbered list
2. Another item
3. One more

[Link text](https://example.com)

That's it. That's 95% of markdown. You now know enough.

Module 1 Capstone — Ship One Real Edit

You now have:

• Claude Code Desktop installed
• A mental model for the three tabs (Chat / Cowork / Code)
• The ability to run a Code session on a real folder
• The diff view workflow — review, accept, reject, comment
• Three steering moves — interrupt, attach, @mention
• Enough markdown to produce any file in this course

Pick one small, real task from a project you actually care about. Something you've been putting off. A typo you've been meaning to fix. A README that needs a closing paragraph. A folder of notes that needs a top-level index.md.

When it lands, you've crossed the line from "user" to "operator." This is the floor. Everything from Module 2 onward makes this floor more powerful.

Next stop: Module 2 — Control Claude. You'll learn the five permission modes (Ask, Auto-Accept, Plan, Auto, Bypass) and pick the right gear for every kind of task. No more constant approval prompts when you trust the work. No more runaway AI when you don't.