What even is Claude Code?
The thing nobody explains in plain English.
You already know Claude in the browser — you type, it types back. Claude Code is the same Claude, but it can actually do things on your computer. It lives in a little black window called the Terminal, and instead of just talking about a task, it can read your files, write new ones, run things, and build stuff for you — start to finish.
Think of the difference like this: chat Claude is a brilliant assistant on the phone giving you advice. Claude Code is that same assistant sitting at your desk, hands on the keyboard, actually doing the work.
The honest bit
Yes, it's a black terminal window and that looks scary the first time. But you never type code into it — you type sentences, the same way you talk to Claude now. The window is just where it works. That's the whole leap.
Which Claude tool do I use?
The number one question I get. Here's the no-second-guessing version.
There are three flavours of Claude and people freeze trying to pick. Don't overthink it — here's what each one is for:
💬 Claude.ai
The browser/app version. Type, get an answer, copy it out.
Best for: thinking, writing, planning, quick questions, one-off tasks.
🪟 Cowork
A friendlier app where Claude works inside folders on your Mac — no terminal.
Best for: organising files, running a workspace, when the terminal feels like too much.
⌨️ Claude Code
The terminal version. Most powerful — it builds, edits, and runs real things.
Best for: building tools, processing data, automating jobs, anything multi-step.
The one-line rule
Need to think or write? → Claude.ai. Need Claude to actually build or run something? → Claude Code. Want the power of Code but the terminal scares you? → start in Cowork. You can use all three — most days I do.
Want the longer breakdown? I made a whole page on it: Claude.ai vs Cowork vs Claude Code vs n8n →
The Tiny Agency mental model
This is the thing that made it click for me.
Here's how I actually think about Claude Code, and it changed everything: you're not the worker any more. You're running a tiny agency, and Claude is your team.
You don't open the laptop to do the task. You open it to brief someone on the task. Your job shifts from "doer" to "delegator" — you say what you want and what done looks like, and the work happens.
You're the founder
You set the direction and decide what matters. You don't need to know how the work gets done — just what you want at the end.
Claude is your team
It does the actual building. Give it a clear brief and it gets on with it — and it can pull in extra helpers (sub-agents) for big jobs without you lifting a finger.
CLAUDE.md is the staff handbook
One file where you write down how your "agency" works — your preferences, your tools, your rules. Claude reads it before every job so you stop re-explaining yourself. (More on this in Section 5.)
Why this matters for you
Non-technical founders get stuck thinking they have to understand the code. You don't. A founder doesn't write the contracts and design the logo and do the books — they hire it out and check the result. Treat Claude Code exactly the same way and it suddenly feels doable.
Where to actually begin
Skip the overwhelm. Here's the exact starting path.
Install it once
Open the Terminal app and paste the install command from Anthropic's setup page. It's one line, you run it once, and you're done. (If that sentence made you nervous — this is exactly what we walk through together in the membership.)
Make a folder for the job
Claude Code works inside a folder. Make one for what you're doing — "content-tools", "client-stuff", whatever — and point Claude Code at it. That folder is its workspace.
Start small and real
Don't try to build an empire on day one. Pick one annoying job — "rename these 40 files", "turn this transcript into a summary", "build me a simple page". Watch it work. That first win is everything.
Talk to it like a person
No magic words. "Can you read this file and tell me what's in it?" works perfectly. If it does something you don't like, just say so — it adjusts.
👉 Do this first
Before you build anything ambitious, do one tiny real task and let it finish. The confidence from seeing it actually work is worth more than reading ten more guides.
Plan Mode + CLAUDE.md memory
The two habits that keep it from going rogue.
Plan Mode — make it think before it touches anything
Left to its own devices, Claude Code will start changing files immediately. Sometimes that's fine. For anything that matters, you want it to plan first, act second.
Press Shift + Tab to flip into Plan Mode (or just type /plan). Claude lays out exactly what it's going to do — every step — and waits for your yes before touching a single file. You read the plan, and if step 3 looks wrong, you catch it before it happens instead of cleaning up after.
Why this is the #1 beginner habit
90% of the "wait, why did it do THAT?!" moments disappear when you plan first. It costs you 30 seconds and saves you from a 40-minute mess. Use it on anything you'd be annoyed to have done wrong.
CLAUDE.md — the memory file that stops the repeating
Claude Code reads a file called CLAUDE.md at the start of every session. It's your project's rulebook — the staff handbook from the Tiny Agency model. Whatever you put in there, Claude knows automatically, every time, without you saying it again.
Drop your preferences in: how you like things written, which tools you use, Australian spelling, "always explain it simply, I'm not a developer" — whatever you find yourself repeating. To add a rule on the fly, just start a message with # and it saves straight into the file.
The rule of thumb
If you've corrected Claude on the same thing more than twice, that's not a Claude problem — it's a CLAUDE.md problem. Write the rule down once and you'll never have that argument again.
Set it loose: /loop + /goal
The two commands that let Claude work while you walk away.
Once you're comfortable, these are the fun ones. Both let Claude keep working without you babysitting every step — they just do it in two different ways. Think of it as the difference between "check on this every so often" and "don't stop until it's done".
What it does: you give Claude a task and a timer, and it re-runs that task on its own — without you asking again. You're telling it two things: what to do and how often.
That runs every 20 minutes in the background of your session. It's perfect for keeping an eye on something — checking a job's progress, watching for a result, nudging you to do a thing later. Units are simple: s seconds, m minutes, h hours, d days. Leave the time off and it defaults to every 10 minutes.
Good to know
A loop only runs while Claude Code is open — close the window and it stops. It also auto-expires after 7 days, so a forgotten loop can't quietly run forever. Tidy and safe by design.
What it does: you give Claude a finish line, and it keeps working — turn after turn — until it genuinely gets there. Normally Claude does one round and hands back to you. With /goal, it doesn't stop and ask; it just carries on.
The clever part: after each round, a second little Claude checks your finish line and decides — are we actually there yet? If not, it tells the main Claude to keep going (and gives it a nudge on what's still missing). You walk away; it grinds through the whole thing.
The catch
/goal needs a finish line it can actually check. "Make it nice" won't work — it can't tell when "nice" is reached. "Every item on this list has a description" works, because that's either true or it isn't. Give it something countable.
The simplest way to remember the difference
/loop = "do this again and again, on a timer." · /goal = "don't stop until this one thing is finished." One repeats forever; one runs to the finish line and stops.
Ready to actually use Claude Code?
This kit is where you start. These are where you go next.
Wright Mode Membership
Where we install Claude Code together, build real tools on live calls, and you've got a community of women doing the same thing — not just saving this for later.
The Codes Cheat Sheet
Once you're in, grab my 12 most-used Claude Code shortcuts — the slash commands and tricks that save you hours and a heap of tokens every single day.
Claude Code Masterclass
The full deep dive. Build automations, process data, and create real tools with Claude Code — no developer experience needed, all in plain English.