What is Claude Code, and what is it actually for?

When people first hear "Claude Code," they usually picture a chatbot with a different name. It is not that. The word Claude gets attached to three different things, and pulling them apart is the cleanest way to understand what this one does.

The first Claude is the chatbot at claude.ai, the box you type into in a browser. The second is the API, which is how other software talks to Claude behind the scenes. The third is Claude Code, which runs in a terminal window on your own computer, can see whatever folder you point it at, and can build, change, and run files there. That third one is the version with hands.

What it's for

Claude Code is for the long list of small useful things you'd otherwise pay a freelancer for or quietly give up on. A simple website that lives at a real address. A quoting calculator for your service business. A script that takes the messy export from your accounting tool and turns it into the clean summary your accountant actually wants. A tiny internal tool that does one job for your team.

It is also for prototypes. If you have an idea and you'd like to see whether it's worth pursuing, Claude Code can give you a working version in an afternoon. If you later hire a real developer for the production version, you'll have something concrete to hand them, which is the most valuable thing a client can bring to a developer.

It is not for the production software that runs your whole business. Where real money moves through code, where sensitive customer data lives in it, where many users hit it at once, you want someone whose full-time job is making that work. Claude Code makes you a smarter client of those people, not a replacement for them.

How it feels in practice

You install it once, sign in with your Anthropic account, and from then on you start it by typing "claude" inside any folder you'd like to work in. A blinking cursor appears. You describe what you want. Claude proposes changes, shows you a before-and-after view of each file, and asks permission before saving anything. You either approve or you correct. When the work is in good shape, you tell it to save a checkpoint, which it does with a one-line commit message you can always return to.

The whole experience is closer to working with a fast, literal, slightly over-eager contractor than to writing software. Most of the craft is in describing what you want clearly, the same skill you already use when you brief a designer or hand off work to a coworker.

What it costs

Claude Code is not on the free tier of Claude. It requires a paid Anthropic plan. For almost everyone, the right one is the Pro plan, around twenty dollars a month, which covers both the chatbot and Claude Code from a shared pool of usage. The heavier Max plans exist for people who use Claude all day. For the kind of projects this category is good for, the Pro plan is plenty.

There is also a metered, pay-per-use option through a separate Anthropic Console account, billed by the token. It is variable rather than fixed, and it adds up quickly if you forget to watch it. For most readers the flat subscription is the simpler, safer choice.

What makes it different from a chatbot

Two things, both important. The first is that Claude Code can see your files. When you ask it to change the look of a button, it does not need you to paste the relevant code. It opens the file itself. That alone removes most of the friction that makes coding with a browser-based chatbot exhausting.

The second is that Claude Code can run things. It can install a tool, start a server, execute a small script, and read what came back. When something breaks, it can usually look at the error, fix the cause, and try again, all on its own. The conversation feels the same. The reach is bigger.

That reach is the whole point. The chatbot can advise you. Claude Code can act. Once you've worked with it for a day, the gap between those two things turns out to be most of the value.