Get one useful result from your own work
Claude Code can work directly with files on your computer. This guide gets it running, then gives it one task you already know how to judge.
For anyone new to Claude Code. By the end you'll have it running and one finished output from your own work. Prerequisite: a Claude Pro, Max, Team or Enterprise plan, or a billed Console account, plus a laptop you can install software on. The free Claude plan does not include Claude Code. Hands-on time: about 30 minutes.
The one idea that makes this click
Claude Code is a version of Claude that can read your files, edit them and run computer commands. Give it the task as you would brief a sharp analyst, then review the work before you use it. The judgment stays with you.
You brief it in plain English. Claude Code handles the computer commands and shows you what changed before you use the result.
Today's scope is installation and one useful result. Coding, git, automations and connectors can wait.
Step 1Install · 5 min
Anthropic's Desktop app can run Claude Code in a graphical window. This guide uses the terminal so you can see each instruction and where your files live.
A terminal is an app where you type an instruction instead of clicking a button. Open it now. (Mac: Cmd+Space, type "Terminal", Enter. Windows: Start, type "PowerShell", Enter.) A window with a blinking cursor appears — that's it. Paste the line for your machine and press Enter:
curl -fsSL https://claude.ai/install.sh | bash
irm https://claude.ai/install.ps1 | iex
brew install --cask claude-code
The native installer updates Claude Code automatically. Homebrew installations update with brew upgrade claude-code.
Commands checked against Anthropic's setup guide on 22 August 2026.
claude: command not foundright after install — close the terminal window and open a fresh one. The install adds Claude to your path, but only new windows pick it up.- Never used a terminal? The official
terminal-configguide and the five-minutequickstartat code.claude.com walk it step by step.
Step 2Log in · 2 min
Type claude. The first time, it opens your browser to log in. Use your Claude plan account, or follow the
Console billing route if that is how you access Claude Code. You'll land in Claude Code: a prompt waiting, your model and folder shown above
it. Type /help any time to see what's available.
Step 3Point it at real work · 8 min
Put a real file somewhere easy and take Claude to it. Pick whatever's closest to your job — a messy CRM export, a folder of last quarter's QBR docs, a spreadsheet of rep numbers you keep re-formatting by hand. Make a folder, drop the file in:
cd Documents/rev-work claude
cd means "change directory" — go into that folder. It's the only terminal command you need today.
Now just ask, in plain English. Real examples — pick the one that fits your file:
what's in this folder? summarise each file in one line. this CSV is a messy export. tell me what columns it has and what looks wrong with the data. read these three QBR docs and give me the five themes that show up in all of them.
Claude reads the files itself — you don't paste anything. Sit with how different that is from chat: it's working on your data, in place.
how does this comp model work? what feeds the bonus column, and where could it break? this is last year's territory file. why is the East quota so much higher than the others? walk me through it before I change anything.
A system explained to you in two minutes instead of an afternoon of spelunking. That alone is worth the install.
Step 4Get one finished output · 7 min
Asking is nice. Making is the win. Ask it to produce something and watch the approval step:
clean this CSV: trim the whitespace, fix the date format to YYYY-MM-DD, and drop rows with no account name. save it as accounts-clean.csv.
When Claude changes or creates a file, it shows you exactly what it'll do and asks before doing it. Nothing happens
to your files without your yes. Read the diff, approve it. Open accounts-clean.csv — it's there, on your disk,
done. That's the loop you'll use forever: describe → it proposes → you approve → it's real.
The one habit to start with: be specific
The single biggest difference between people who love this and people who bounce off it is specificity:
- Weak: "clean this up."
- Strong: "the Amount column has £ signs and commas so it's reading as text — strip those, make it a number, and tell me the new total."
You brief Claude the way you'd brief a good analyst. The more precise the brief, the less you correct.
- It can be confidently wrong. It'll produce a clean-looking total that's off if the data tricked it. Always have a way to check. Never ship a number you didn't verify.
- It works on what you point it at. Garbage export in, garbage out. Give it the real file, not a screenshot.
- Long sessions get worse, not better. If it drifts, type
/clearand start the next task fresh. - Not "set and forget" yet. Today you're driving. Automation comes later.
- Find one file you'd normally clean, reformat, or summarise by hand.
cdinto its folder, runclaude, ask for the finished version.- Approve the change. Open the result. Save the exact wording that worked — you'll reuse it. That's the start of your own playbook.