Oleg MelnikovYouTube
Claude Code

Claude Code, for people who are not developers

I am not an engineer any more, i run marketing, and Claude Code is the thing i use all day. Here is what it is, how to install it, and the four ideas that make it click. No codebase required.

Get Claude Code

Setup guide

Pro at $20 a month is the way in. The free plan does not include Claude Code, which is why the install can work and the login still refuse you.

What Claude Code is, in plain language

Claude Code is Claude with your files open. It runs in a terminal, it works inside whatever folder you start it in, and instead of telling you what to do it goes and does it, then checks whether what it did worked.

The name is the problem. It says code, so everyone who is not a programmer assumes it is not for them, and that is wrong in a way that costs people a lot. Most work that is not coding is also files and commands. A folder of transcripts. A spreadsheet of leads. Two hundred images that need renaming. A script that calls an API and writes the result somewhere. All of that is exactly what it is good at.

Everything on this site was built with it by someone doing marketing. The barrier is being willing to open a terminal, not knowing how to program.

Installing it, on any machine

What you need
A paid Claude plan from $20 a month, or an API account
Operating system
Macos 13+, windows 10 1809+, ubuntu 20.04+, debian 10+, alpine 3.19+
Hardware
4 GB of ram, x64 or arm64
Time
About five minutes, most of it waiting
Also available as
A desktop app, in vs code and jetbrains, on the web, in slack, and in ci

From Anthropic's setup docs, read 2026-08-27.

  1. 1Get a plan

    Claude Code is not free. You need a paid Claude plan (Pro at $20 a month, Max, Team or Enterprise) or a Claude console account with credit on it.

    This trips up almost everyone once, because the install itself works perfectly and then the login refuses you. Nothing is broken. The full cost picture is on The pricing page.

  2. 2Install it with one line

    Open a terminal and paste the line for your machine.

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

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

# Windows CMD
curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

    Not sure which windows shell you are in? Your prompt shows PS C:\ in PowerShell and C:\ in cmd.

    Homebrew (brew install --cask claude-code) and winget (winget install Anthropic.ClaudeCode) both work too, with one catch worth knowing: the native install updates itself in the background, and the homebrew and winget ones do not. If you install that way you have to run the upgrade yourself.

  3. 3Close the terminal, open a new one

    This is not optional and it is the reason half the "command not found" comments exist. The terminal reads your PATH when it starts, so the window that ran the installer does not know about the new command yet.

    In the new window, claude --version should print a number followed by (Claude Code).

  4. 4Start it in a folder that matters

    Claude Code works inside the folder you launch it from. So go somewhere real first, then type claude.

    If you are not a developer and have no code to point it at, make an empty folder for the thing you actually want. I started mine as a folder with a text file describing my content process, and everything since has grown out of that.

  5. 5Ask before you tell

    The first prompt should be a question, not a job. "what is in this folder and what does it do?"

    Two reasons. You find out how it explains things, which tells you how much context it needs from you. And if it is wrong, nothing has happened yet.

  6. 6Write down how you work, once

    Make a file called CLAUDE.md in the folder. Put in it what this project is, how you want things done, and anything you have found yourself saying twice.

    Claude Code reads it at the start of every session. This is the single highest-leverage thing on this page: it is the difference between a clever assistant you re-brief daily, and something that already knows how you work.

Your first twenty minutes

A sequence that works, in the order that gets you to something real fastest.

  1. Ask it what is in the folder. You are testing how it explains, not what it knows.
  2. Ask it to make one small change you could check by eye in ten seconds. Rename something. Fix a typo across five files.
  3. Ask it to do something you would have to look up. Convert a folder of images. Pull the numbers out of a CSV into a summary. This is where it stops being a demo.
  4. Write your CLAUDE.md. Five minutes, and every session after this one is better for it.
  5. Give it something real and walk away. Come back and read what it did.

The one habit worth forming on day one

Read what it changed before you accept it. Not forever, just until you have a feel for where it is reliable and where it guesses. That calibration is the whole skill, and you cannot get it by reading about it.

The commands that actually matter

There are a lot. These are the ones I use every day, and the rest you can meet when you need them.

What it does

/clear
Throws away the conversation so far
/usage
Shows what this session and this month have used
/model
Switches which model is answering
Plan mode
Makes it write the plan before touching anything
/doctor
Prints a health report on the install
/help
Lists everything else

When I reach for it

/clear
Every time I switch to an unrelated task. Keeps answers sharp and the bill small.
/usage
Once a week, and any time something feels slow
/model
Down to a cheaper one for routine editing, up for anything that needs judgement
Plan mode
Anything that will change more than a couple of files
/doctor
First thing when something is broken and I do not know why
/help
More often than I expected to

The four ideas that make it click

People who get value out of Claude Code and people who bounce off it are usually separated by these four, not by technical skill.

1. Context is the product. The quality of what you get back is decided almost entirely by what it can see. Naming the three files that matter beats letting it search the whole folder, and it is cheaper. Carrying an old conversation into a new problem makes the answers worse, not better, which is what /clear is for.

2. CLAUDE.md is where you stop repeating yourself. A file in the folder that it reads every session. What this project is, how you want it done, the rules that are not obvious. Anything you have explained twice belongs there instead.

3. Plan first on anything that matters. Ask for the plan before the work. You get to correct a paragraph instead of a day of edits, and it is the cheapest review you will ever do.

4. It is a delegation habit, not a typing habit. The failure mode is treating it like autocomplete and staying in the loop for every keystroke. The win is handing over a whole job, going and doing something else, and reviewing the result. That feels wrong for about a week and then it does not.

Past those four there are skills (reusable instructions it loads on demand), sub agents (separate workers for separate parts of a job), MCP servers (connections to outside services) and hooks (things that run automatically at certain points). None of them matter until the four above are habits.

What I build with it

Every guide below is a real system I run, written out in full so you can rebuild it without watching anything. They are the answer to "fine, but what do I actually do with it".

And if the job is a document rather than a repository, Claude Cowork is the other half of the same plan.

The mistakes everyone makes first

  • installing the VS Code extension and expecting the claude command. The extension keeps a private copy for its own panel. Install Claude Code itself.
  • Not opening a fresh terminal after installing. The old window does not know about the new command.
  • Expecting the free plan to work. It installs and then will not sign you in, which reads as a bug and is not one.
  • Letting one session run all day. The conversation becomes input on every turn, so it gets slower, more expensive and less accurate at the same time.
  • Skipping permission prompts to go faster. The prompts are the only thing standing between a misunderstanding and a deleted folder. Read them until you have a feel for it.
  • Never writing a CLAUDE.md. The most common one, and the most expensive.

The errors people actually hit, taken from the comments under my videos, are answered in the troubleshooting section above. What it costs, including Anthropic's own figures for what real usage runs to, is on The pricing page. And if you are still deciding between this and an editor, the Comparison is next door.

Install commands and system requirements checked against Anthropic's own quickstart on 2026-08-27.

If you get stuck

The things that actually go wrong, taken from the questions people ask under the video.

The terminal says “command not found: claude”

Installing the Claude Code extension in VS Code does not give you the Claude terminal command. The extension keeps a private copy for its own chat panel. That catches almost everyone, and it is not a broken setup.

To get the command, install Claude Code itself:

# macOS or Linux
curl -fsSL https://claude.ai/install.sh | bash

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

Then close the terminal window completely and open a new one. Skipping that is the second most common reason it still says not found. Check it worked with Claude --version.

Still not found on a Mac? Your PATH is missing the folder, so run these two lines:

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc
source ~/.zshrc

Claude doctor prints a full health report if you are still stuck. And if you would rather never touch a terminal, the Claude Code desktop app does the same job with buttons.

One thing worth knowing before you install anything: Claude Code needs a paid Claude plan (Pro or higher) or an API account. The free claude.ai plan does not include it, so it can install perfectly and still refuse to sign you in.

What does this actually cost to run?

Claude Code comes with a Claude subscription: Pro at $20/month, or Max from $100/month for five or twenty times the usage. Usage is included, there is no per-run charge. The free plan does not include Claude Code at all.

Inside Claude Code, /status tells you whether you are on a subscription or an API key, and /usage shows how close you are to your plan limit.

Prices and free tiers move. These were checked in August 2026.

Is --dangerously-skip-permissions safe?

It does exactly what the name says: it turns off the ask-before-acting prompt, so Claude can edit files and run commands on your machine without checking with you each time. That is why it is fast, and that is the whole risk.

The honest version: in a project folder like this one, where you know what is inside and nothing valuable is nearby, it is a reasonable trade. Pointed at your whole home folder, or at a repo you have not read, it is not.

The safer default is to drop the flag. Claude "/prime" does the same work and asks first. You approve a few things at the start, then it stops asking about the same kinds of actions.

It says “credit balance too low, add funds”

That message is about the Anthropic API, which is prepaid pay per token. It is not about your Claude subscription, and a Pro or Max subscription never funds an API key. They are two different products with two different bills.

If you meant to use the API, add credit at platform.claude.com under Billing, and switch on auto reload so a long run cannot die halfway.

If you did not mean to use an API key, you probably have one sitting in your environment, and a key set there overrides your subscription login. Check with echo $ANTHROPIC_API_KEY on a Mac. If it prints something, remove it and run /login inside Claude Code to sign in with your subscription instead.

Questions people ask

What is Claude Code
Claude Code is Anthropic's agent that runs in your terminal and works directly on the files in a folder. Instead of answering a question in a chat window, it reads your files, edits them, runs commands and checks whether what it did worked. It is also available as a desktop app, in VS Code and JetBrains, on the web, in Slack, and in CI through GitHub Actions.
Is Claude Code only for developers
No, and that is the most useful thing to know about it. It edits files and runs commands, and most work that is not coding is also files and commands: spreadsheets, transcripts, folders of images, scripts that call an API. Everything on this site was built with it by someone doing marketing, not engineering. The barrier is being willing to use a terminal, not knowing how to program.
How do I install Claude Code
One line. On macOS, Linux or WSL run curl -fsSL https://claude.ai/install.sh | bash. In Windows PowerShell run irm https://claude.ai/install.ps1 | iex. Then close the terminal completely, open a new one, and run claude --version to check. Homebrew and WinGet also work but do not auto-update.
Why does it say command not found Claude
Almost always one of two things. Either you did not open a fresh terminal after installing, so the window does not know about the new command yet, or you installed the VS Code extension rather than Claude Code itself: the extension keeps a private copy for its own panel and does not give you the terminal command. On a Mac you may also need to add $HOME/.local/bin to your PATH.
What is the difference between Claude Code and Claude chat
Chat answers you and you do the work. Claude Code does the work. It has your files, it can run commands, and it checks its own results, so the output is a changed folder rather than text you have to copy somewhere. If the job ends with you pasting an answer into something, chat is enough. If it ends with a file being different, that is Claude Code.
What is a CLAUDE.md file
A plain markdown file in your project folder that Claude Code reads at the start of every session. You put in it what the project is, how you want things done, and anything you have caught yourself explaining twice. It is the highest-leverage five minutes you can spend, because it turns a general assistant into one that already knows your rules.
Is Claude Code better than cursor
They are different shapes rather than better and worse. Cursor is an editor you work inside, so it suits writing code by hand with help. Claude Code is an agent you delegate to, so it suits handing over a whole task and reviewing the result. For non-coding work like content pipelines and research, Claude Code fits better because the work is folders and commands rather than a codebase in an editor.

Claude Code will build you the system. It will not tell you what you should be known for. Boldane does that half for founders with real expertise, from one hour of talking a week.