Skip to content

What Is Claude Code? A Plain-English Guide for 2026

What is Claude Code? A plain-English guide to Anthropic's AI agent: what it can touch on your computer, what it costs and 6 business tasks it handles well.

By Yotam DahanUpdated 10 min readClaude Code

On this page
  1. How Claude Code works, in plain English
  2. Claude chat vs Claude Code vs the Claude desktop app
  3. Where you can run it
  4. What Claude Code can touch on your computer
  5. What it costs to use
  6. 6 business tasks Claude Code handles well
  7. When not to use Claude Code
  8. What to do next

Claude Code is Anthropic's AI agent that works directly in a folder on your computer. You describe a task in plain English, and it reads your files, creates and edits documents and code, and runs small programs until the task is done, showing you each change along the way. It is included in every paid Claude plan, starting with Pro at $20 a month.

If you have only used Claude in a browser chat, the difference is simple. Chat gives you text that you copy, paste and save yourself, while Claude Code does the copying, pasting, saving and checking for you.

How Claude Code works, in plain English

Claude Code is an agent: an AI that can use tools to take actions, not only write replies. In practice, that means it reads your project, edits files and runs commands in a repeating loop.

One Claude Code task, start to finish
  1. Your goal

    You describe the finished result in plain English

  2. Look around

    It lists the folder and opens a few files to see what is there

  3. Act

    It writes files, runs small programs or searches the web

  4. Check

    It compares the result with your request and fixes the gaps

  5. Report

    It sums up what it did and waits for your next instruction

A goal can be as ordinary as this one:

A typical first goal
Turn the [NUMBER] supplier PDFs in this folder into one spreadsheet with date, supplier, amount and VAT. List any file you couldn't read.

You can interrupt at any moment with Esc, or type a correction while it works and it reads it before its next step. That loop is why it feels less like a chatbot and more like a quick assistant at your keyboard, one who still needs a clear brief and a second pair of eyes.

Two words you will meet often. The terminal is the text window where you type commands (Terminal on a Mac, PowerShell on Windows), and a session is one conversation with Claude Code about one folder.

Claude chat vs Claude Code vs the Claude desktop app

People mix these up all the time, partly because they share a name and an account. Here is how they differ.

Claude chatClaude CodeClaude desktop app
What it isA conversation with ClaudeAn agent that works on the files in a folder you chooseOne installed app that holds chat and a Code tab
Where you use itclaude.ai, the mobile app, the desktop appThe terminal, the desktop app's Code tab, VS Code, JetBrains IDEs, or claude.ai/codeMac and Windows, with a Linux version in beta
Edits files on your computerNoYesPartial
Runs commands on your computerNoYesPartial
What you get backAnswers, drafts and files you downloadFinished files saved straight into your folderBoth
Best forQuestions, writing, thinking a problem throughMulti-step work across many files, building things, repeat jobsUsing Claude Code without touching a terminal
Included on the free planYesNoPartial

In the desktop app, "Partial" means the chat side works like Claude chat, while the Code tab is Claude Code itself: you pick a folder and Claude works on its files.

Where you can run it

Four ways to open Claude Code

Terminal

The full-featured version. You install it with one command, then type claude inside any folder.

Desktop app

Open the Code tab, pick a folder and type your request. You review changes visually, no terminal needed.

Code editors

An extension for VS Code and Cursor, and a plugin for JetBrains editors such as IntelliJ and PyCharm.

Browser

At claude.ai/code, sessions run in the cloud on a GitHub repository, which suits code projects best.

All four run the same Claude Code engine, so a project's instructions and settings work in each of them. If you have never opened a terminal, the desktop app is the gentlest start, and our step-by-step install guide covers both routes on Mac and Windows.

What Claude Code can touch on your computer

This is the part people worry about most, so here is exactly what it can reach once you start it in a folder:

  • That folder and everything inside it. It reads files there freely. Reading or writing outside that folder triggers a permission request, unless you deliberately add another folder.
  • Your terminal. It can run any command your user account could run, such as moving files, converting images or running a script. Whether it asks first depends on the permission mode.
  • The web. It can search and open web pages when the task needs fresh information.
  • Tools you connect. Through MCP connectors (MCP is the open standard that lets AI tools plug into other apps), it can use services such as Google Drive, Gmail or Slack, but only the ones you set up yourself.
What keeps it in check

Permission modes

Manual asks before every edit and command. Auto, the starting mode in current versions, has a second AI model review each action and block risky ones.

Folder trust

The first time you open a folder, Claude Code asks whether you trust it before it loads that folder's settings.

Undo for file edits

It snapshots files before editing them, so you can rewind. Changes made by commands it ran, and anything sent to outside systems, aren't covered.

Local history

Sessions are saved as plain-text files in a hidden .claude folder in your home folder for 30 days by default, which is how resuming works.

Auto mode blocks actions such as downloading and running unknown code or deleting files that existed before the session; Anthropic's permission modes page lists every rule. You switch modes with Shift+Tab, and the data usage page explains what is stored where.

When I set this up for a client, the first rule is always the same: give Claude Code its own project folder with copies of the files it needs, not the whole hard drive.

What it costs to use

Claude Code is free to download, but it only works with a paid account. You either use a Claude subscription or pay per use through Anthropic's developer console.

Monthly price per person in USD, billed monthly, October 2026
  • Pro$20
  • Team Standard seat$25
  • Max 5x$100
  • Team Premium seat$125
  • Max 20x$200

From the official pricing page and Anthropic's Max plan help article, checked October 2026.

  • Free plan: $0, but Claude Code isn't included.
  • Paying yearly: Pro drops to $17 a month ($200 up front), and Team seats drop to $20 and $100.
  • Enterprise: $20 per seat a month billed yearly, with a 20-seat minimum, plus usage at API rates.
  • API through the Claude Console: no subscription, you pay for every token you use.

A token is a small chunk of text that the model reads or writes. On subscription plans your usage runs on a rolling five-hour window plus a weekly cap, and your chats and Claude Code sessions draw from the same pool.

Heavy sessions on large folders use that pool faster than a few chat questions do. The guide to Claude Code usage limits shows how to check and stretch it, the Claude Code pricing guide compares plans by type of user, and prices change, so check the official page before you buy.

6 business tasks Claude Code handles well

These are the jobs where an agent beats a chat window, because they involve real files and several steps:

TaskWhat you give itWhat you get back
Market research briefA list of competitors or a question about your marketA written brief saved as a file, with links to every source
Spreadsheet cleanupA messy CSV export from your CRM, shop or bankA cleaned file plus a short note of what it changed
PDFs into dataA folder of invoices, quotes or receiptsOne spreadsheet with the columns you asked for
A landing pageYour business details and a style you likeA working page you can open in your browser and later publish
Documents from rough notesMeeting notes, a voice-note transcript or an outlineA proposal, procedure, FAQ or set of email drafts in your tone
Small repeat jobs"Every Monday, do this with this export"A script or a scheduled task that repeats the job for you

For 15 more examples grouped by job, each with a prompt you can paste, see Claude Code for non-developers.

When not to use Claude Code

It is a powerful tool, which is exactly why it is the wrong one for some jobs:

  • Quick questions and one-off writing. For "rewrite this email" or "what does this clause mean", the chat app is faster and uses less of your plan.
  • Live systems where a mistake can't be undone. Payments, customer databases, a live online store or anything that messages customers. Let it prepare the change or the draft, and do the final click yourself.
  • Decisions you can't check. Tax, legal, medical and investment calls need a qualified person. Use Claude Code to collect, organize and summarize the material, not to make the call.
  • Data you're not allowed to share. If a contract or a regulation says client data can't go to outside services, that applies here too. On Free, Pro and Max plans, review your model-training setting first.
  • Simple app-to-app connections. If all you need is "every new form entry goes into a spreadsheet", a no-code automation tool is usually simpler to run. Our explainer on what AI automation is helps you choose.
  • The free plan. Claude Code is not included, so stay in chat until you have a task that justifies a paid plan.

Quick check0 of 3 correct

What can Claude Code reach by default when you start it in a folder?

What is the cheapest way to get Claude Code?

Which of these can Claude Code's rewind not undo?

What to do next

Install Claude Code, create an empty test folder, and ask it to explain what it can do there before you let it change anything. The install guide walks you through Mac and Windows, and the beginner tutorial takes you through a first real project over a weekend. If you want a structured path from first command to finished projects, the Claude Code course covers setup, CLAUDE.md and prompting, then builds a landing page, an AI video and an invoice agent with you.

Chat, Claude Code or something else? Task decision sheet

A one-page table that sorts 20 common business tasks into chat, Claude Code, no-code automation or a human, plus 5 questions to ask before you delegate.

Enter your email and the full section opens right here. You also get a free lesson from the course.

No spam. One click to unsubscribe. We use your email to send you the course and AI tips.

Frequently asked questions

Who makes Claude Code?

Anthropic, the company behind the Claude models and claude.ai. One Claude account and one paid plan cover both the chat app and Claude Code.

Which AI model does Claude Code use?

It runs on Anthropic's Claude models. On Pro, Max, Team and Enterprise plans the default is currently Opus 5.5, and you can switch to another model, such as Sonnet or Haiku, with the /model command.

Does Claude Code work without an internet connection?

No. The model runs on Anthropic's servers, so Claude Code needs a working internet connection the whole time, even though your files stay on your computer.

Can I use Claude Code from my phone?

Partly. The Claude app for iOS and Android acts as a remote: it can start and watch cloud sessions or continue a session that is running on your computer, but the work itself never runs on the phone.

Does Anthropic train its models on what I do in Claude Code?

On Free, Pro and Max plans it depends on a privacy setting you control in your claude.ai settings. On Team, Enterprise and API accounts, Anthropic does not train on your Claude Code prompts or code unless your organization opts in.

Keep reading