What Claude Code actually is
Claude Code is a program from Anthropic that puts Claude — the AI model — to work on your computer. You describe what you want in ordinary English: "build a page that lists my invoices", "find out why this script crashes", "rename every file in this folder by date". Claude reads your files, writes the code, runs it, checks its own work, and reports back.
The "agent" part is what separates it from a chat window. A chatbot hands you text to copy and paste. Claude Code does the work itself, on your machine, in your actual project — and, by default, asks your permission before each action. The reason it feels different is simple: it doesn't describe the work. It does it.
So what is "the terminal", and why does it feel hostile?
The terminal is the oldest way of talking to a computer: a window where you type a command, press return, and read the reply. No buttons, no menus, no icons. It predates the mouse, and it never picked up the habits that make modern software forgiving — no visible undo, no tooltips, no hint about what to type next.
That's why it feels hostile. It isn't that you missed a lesson everyone else attended; it's that the terminal assumes knowledge rather than offering it. A blank prompt gives you nothing to go on, and a mistyped command answers with an error written for the person who built the tool. The problem is the interface, not you. Programmers keep using it because it's fast and direct once memorised — but "once memorised" is doing a lot of work in that sentence.
The three things that trip people up
In practice, almost everyone gets stuck at the same three points.
- Installing it. Claude Code is installed by typing a command into the terminal — so the tool begins with exactly the kind of task it's meant to help with. If your Mac is missing a piece the installer expects, the error message will be accurate and unhelpful in equal measure.
- The permission prompts. By default, Claude asks before it acts: "Can I run this command?" That's the right design — but the question arrives in the language of shell commands, and it's hard to approve something you can't read. Most people either freeze or approve everything, and neither feels good.
- Losing the thread. Sessions get long. You close the window, come back tomorrow, and the conversation where Claude finally understood your whole project is buried in a scrolling list. Claude Code can resume old sessions — provided you can find the right one.
How to actually run it
If you'd like to do it by hand, here is the honest version. It's five steps, and none of them bites.
Open Terminal: press ⌘-space, type "Terminal", press return. Install Claude Code with the one-liner below, then move into your project's folder — type cd, a space, and drag the folder onto the window so its path types itself. Type claude and press return. The first time, it opens a browser window so you can sign in with your Anthropic account.
npm install -g @anthropic-ai/claude-code
Anthropic occasionally changes the installer — the current command lives in the Claude Code docs, which are the source of truth.
From there, you type what you want in plain English. Claude proposes actions and waits for your OK. When you're done, close the window; next time, claude --resume lists your past sessions. That's the whole ritual. The terminal is just the room the conversation happens in — and if you'd rather have a better room, read on.
Or skip the scary part
AGI for Anyone is a small Mac app that keeps the parts of the terminal that matter and removes the parts that get in the way. It runs Claude Code in a real terminal — a native PTY, not a chat lookalike — so nothing is watered down. But the setup is one click: the app installs Claude Code for you and detects whether you're already signed in to your Anthropic account. You choose up front whether Claude acts on its own or asks before each step, and you can change your mind at any time in Settings.
The daily friction goes too. Your projects and past sessions sit in a sidebar; quit the app and it reopens where you left off, ready to resume. Press ⌘K and search the full text of every session across every project — that conversation from three weeks ago is one query away. When Claude needs your OK, or when it's your turn, you get a normal macOS notification instead of a window to babysit. The small things are looked after as well: drag a file in and its path types itself, ⌘F searches inside the terminal, sessions can be renamed, and your git branch is visible at a glance.