What is Claude Code? A plain-English guide for business owners

Claude Code is Anthropic's AI coding agent. You describe a job in plain English, and it reads the files in your project, makes edits, runs commands such as a build or a test, checks the result and keeps going until the job is done or it needs your input. Anthropic's own definition is short: "Claude Code is an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools."

The difference from a chatbot is that Claude Code acts. A chat assistant gives you text that you copy into your site yourself. Claude Code works on the files directly, inside limits you set. That is what makes it useful for real website work, and it is also why it needs rules, a copy of your site to work on, and a person who reviews what changes.

This guide explains how it works, what it can reach, what it costs at a glance as of October 2026, and what it is not. If you only need the short version: it is a capable tool that does real engineering work, and people still own the decisions, the facts and the final review.

Claude Code in one paragraph

Think of Claude Code as a fast, well-read junior engineer who sits inside your project folder. You give it a task, such as "add a FAQ section to every service page using the questions in this spreadsheet." It looks around the project to understand how pages are built, makes the changes, runs whatever check you have (a build, a link checker, a test), and reports back. Anthropic's glossary puts it this way: "Claude Code is the harness; Claude is the model inside it." Claude is the AI model that reasons. Claude Code is the software around it that gives it file access, a terminal, permission checks and memory.

How it works: the agentic loop

Anthropic's docs say that when you give Claude a task, "it works through three phases: gather context, take action, and verify results." In practice, a single request can involve dozens of small steps. The How Claude Code works page uses the example "fix the failing tests," where Claude might:

1. Run the test suite to see what is failing.
2. Read the error output.
3. Search for the relevant source files and read them.
4. Edit the files to fix the problem.
5. Run the tests again to confirm the fix.

Each step feeds the next. You can interrupt at any point: pressing Esc stops it, and typing a correction redirects it. The loop is only as good as the check at the end. Anthropic's best-practices page puts the first tip plainly: "Give Claude a check it can run: tests, a build, a screenshot to compare." On a website, that might be a build that must pass, a crawl that must show no broken links, or a before-and-after screenshot of a page.

What Claude Code can touch

When you start Claude Code in a folder, it gets access to a specific set of things. According to Anthropic's docs, that includes:

• Your project files, in that folder and below it, plus files elsewhere if you allow it.
• Your terminal. "Any command you could run": build tools, git, package managers, scripts.
• Your git state: the current branch, uncommitted changes and recent history. Git is the version-control system that records every change so you can see and undo it.
• Your CLAUDE.md file, a plain text file of project instructions that loads at the start of every session.
• Extensions you add, such as MCP servers that connect it to outside tools, and skills that package repeatable workflows.

That list is the reason to be careful. If your terminal can deploy to production or reach a live database, Claude Code can too, unless you block it. Before Claude edits a file it saves a checkpoint you can rewind to, but the docs are clear that "actions that affect remote systems (databases, APIs, deployments) can't be checkpointed." Our article on whether Claude Code is safe for a live website covers the setup we recommend.

Approvals and permission modes

A permission mode sets what Claude can do without asking you first. The modes, as listed on Anthropic's permission modes page in October 2026:

Two points matter for a business. First, with Claude Code version 2.1.283 or later, auto mode is the built-in starting mode for interactive terminal and VS Code sessions. In auto mode, "a separate classifier model reviews actions instead of you." It blocks risky actions, but it is not a person checking your work. If you want to approve every change yourself, start in Manual mode or switch with Shift+Tab. Second, bypass mode skips the checks entirely, and Anthropic warns it "offers no protection against prompt injection or unintended actions." Don't use it on a real site. Anthropic reserves it for isolated containers and virtual machines.

Plan mode is the one most people should use more. Claude researches and writes a plan, and nothing in your source files changes until you approve it. For anything site-wide, read the plan first.

Where you use it

Claude Code runs on several surfaces, and they share the same engine and project settings. As of October 2026:

• Terminal (CLI): the most complete version, and the only one for scripting.
• Desktop app: the Code tab in the Claude desktop app, with visual diff review and no terminal needed. It requires a paid subscription.
• IDE extensions: VS Code (also installs in Cursor) and JetBrains editors.
• Web: cloud sessions at claude.ai/code that keep running after you close your laptop. They clone your code from GitHub and push branches back to it, so connect GitHub first.

Our guide to choosing a Claude Code surface compares them side by side. If you are deciding between Claude Code, the regular Claude chat app and the developer API, read Claude.ai vs Claude Code vs the Claude API.

What it costs at a glance

Claude Code is included in Anthropic's paid plans and can also run on pay-as-you-go API billing. The setup docs say: "The free claude.ai plan does not include Claude Code access." As of October 2026, the individual Pro plan is $20 a month billed monthly, or $17 a month billed annually; Max costs $100 or $200 a month; Team seats, Enterprise seats and pay-as-you-go API billing are the options for organizations.

Pro, Max and Team plans have usage limits that reset every five hours and weekly, shared with your regular Claude chat use. Anthropic publishes multipliers between plans, not exact numbers. Usage-based Enterprise plans have no per-seat limits; usage is billed at API rates. Prices and limits change, so confirm on Anthropic's pricing page before you buy. Our Claude Code pricing guide goes through every plan and what the limits mean in practice.

What Claude Code is not

• Not a website builder. It does not host your site or give you a drag-and-drop editor. It works on code and files you own. If you want a hosted builder, that is a different kind of tool.
• Not a source of truth. It can write a confident sentence about your prices, your licenses or a law that is simply wrong. Facts about your business come from you.
• Not a reviewer of its own work. It can run checks, and it should. A person still needs to read the change before it ships.
• Not a live-site editor. It is best on a copy of your site in git, with changes reviewed and deployed on purpose. Work done only in a live CMS admin, or content stored only in a database, is harder for it to reach safely.
• Not free. It needs a paid plan or API billing, and heavy use costs more.

A real example: eseospace.com

We run our own site with Claude Code. In June 2026 we moved eseospace.com off WordPress to a static build, about 2,227 URLs, and it went live on the new host on June 23, 2026, the day after the first commit. Claude Code wrote the crawler, the publishing pipeline, the deploy steps and the form handler, and applied site-wide edits. People decided what mattered: keeping the original design over a clean rewrite, the host, every deploy, and every brand fact, price and claim.

The first attempt was a clean rebuild that stripped the theme's JavaScript. It broke animations, accordions and menus, and a person abandoned it on day one. That is the pattern to expect. Claude Code does the volume of work, and a person decides what is acceptable. The full story, including the mistakes, is in our eseospace.com rebuild case study.

Who should use it, and who should wait

It suits teams that have their site in code (a static site, a Next.js app, a WordPress theme or a Shopify theme in git) and someone who can read a change before it goes live. It also suits owners who want a repeatable process for bulk work, such as updating hundreds of pages to a new standard. For a list of jobs, see 25 things Claude Code can do on a business website.

Wait if your site lives only inside a hosted builder with no code access, if nobody on the team can check a change before it ships, or if you have no backups or staging copy. Fix those first. A copy of the site, version control and one person who reviews changes will help you more than any tool.

Conclusion

Claude Code is an AI agent that works directly on your project: it reads, edits, runs checks and reports back, inside permission limits you control. It is fast and capable, and it still needs a copy of the site to work on, git to track every change, and a person who approves what ships. Start small, in plan mode or Manual mode, on a task you can easily check. If you would rather have a team set it up or do the work, our Claude Code development service builds and maintains sites this way. More guides are in our Claude Code Guide.

Frequently asked questions