Guides
Guide · 3 min · Anyone

Set up the tools

A blank machine to a working agent, in the order that avoids the wasted hour.

Read this whenYou have heard what these things do and you have not installed one.

The setup guides you have found are wrong in the same way. Eight tools, a comparison table, and a paragraph about how it depends on your workflow. You do not have a workflow yet. That is the problem.

The whole thing is three installs and one habit. About twenty minutes.

Backwards, and the afternoon goes to a path variable instead of to the work.

One. The terminal

You already have one. On a Mac it is Terminal, in Applications then Utilities. On Windows, install Windows Terminal from the Microsoft Store. Do not go and learn the shell first.

Two. The editor

Install VS Code from code.visualstudio.com. You are not going to write code in it. You are going to watch the agent change files and read what it did. Add the Claude Code extension later.

Three. The agent

One line, and it is the same line on Mac, Linux and WSL.

Installing Claude Code
# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bash

# Windows PowerShell
irm https://claude.ai/install.ps1 | iex

# or on a Mac, if you already use Homebrew
brew install --cask claude-code

# then check it worked
claude --version

The native install is the one to use. It keeps itself up to date in the background, which the Homebrew and WinGet versions do not. If claude --version prints a number, you are done installing.

Starting it, in the folder you want it to work on
cd /path/to/your/project
claude

It opens a browser to log you in once, then remembers.

Four. The step everybody skips

Point it at something you already understand and ask what it does. You can tell straight away when the answer is wrong, which is the only way to calibrate what it is good at.

The second agent, later

Once one is normal to you, a second earns its place. Not because it is better. Because it fails differently. Give it a problem the first one is stuck on and read the disagreement.

The overlap is why the second one is cheap to try. The edges are why it is worth keeping.

OpenAI's Codex is the obvious second. Its install command has changed more than once, so take it from OpenAI's own documentation rather than from a guide, this one included.

The one file worth writing

Put a file called CLAUDE.md in the root of your project. What it is, what it is built with, any rule you are tired of repeating. The agent reads it every session.

Five minutes to write, then read back on every session for the life of the project.
CLAUDE.md, and this is genuinely enough to start
# What this is
A personal site. Next.js, TypeScript, Tailwind.

# Rules
- No new dependencies without asking.
- Match the style of the file you are editing.
- If you are unsure, ask one question rather than guessing.

Everything you leave out of that file is something you will type forty times.

What you can ignore for now

  • Every extension, plugin and marketplace. None fix a day one problem.
  • The comparison videos. Benchmarks run on somebody else's codebase.
  • Learning git properly. Three commands, picked up by watching.
  • Choosing a framework. No framework means no project yet, and that is the third guide.

When you can open a folder, start the agent, ask a question about your own code and understand the answer, the setup is finished. A lower bar than the internet suggests, and the actual one.