Claude Code: the coding agent in your terminal (and not just for programmers)
- to get started
- a Claude subscription (Pro or Max) or an API key · terminal, desktop or web
- when to pick it
- You want to delegate real tasks on your computer (code, files, automations) to an agent that works in steps and shows you what it does.
- official site
- code.claude.com
pricing
| plan | checked on | |
|---|---|---|
| Pro | $20/month, includes Claude Code with a session budget suited to short sessions | 19/07/2026 |
| Max | $100 or $200/month, 5x or 20x Pro usage per 5-hour window | 19/07/2026 |
| API | pay per token for the model you choose, no hourly windows | 19/07/2026 |
Claude Code is Anthropic’s agent living in the terminal (and by now also in a desktop app, on the web and inside editors): you describe a task, it reads the files, proposes changes, runs commands and shows you every step before taking it. The difference from a chat is the full cycle: it actually reads your project, acts, verifies the result and corrects course.
The name misleads: code is the home turf, but the job it does is «carry out this task, step by step, in my environment». Many of the automations described in this site’s scaffolding section are built exactly this way, without being a professional developer.
What it costs
Claude Code is included in the Claude subscription: the Pro plan ($20/month) covers contained working sessions, the Max plans ($100 or $200/month) give 5 or 20 times Pro usage per five-hour window, for people who work with it all day. The alternative is a pay-as-you-go API key: you pay for the tokens you use, with no windows, which suits occasional or automated use. Figures checked on 19/07/2026 on the official pricing page; the per-window usage limits change often, so always re-read the exact number there.
How it works
Three properties stand out in daily use. First, explicit permissions: any action outside the agreed perimeter (writing a file, launching a command) asks for confirmation, and you widen the perimeter as trust builds. Second, context files: a CLAUDE.md file in the project gives it the house rules, and skills add reusable procedures. Third, it holds up on long tasks: it plans, executes in steps, and picks up where it left off.
The limits, said plainly
You need some familiarity with the terminal, or the will to acquire it: the desktop app and the web lower the step, but the logic remains «give a task, supervise the steps». The hourly-window plans can squeeze intense days (Max exists for that), and subscription use is tied to Anthropic’s models: if you want to choose the model freely, look at OpenCode. If you live inside the editor rather than the terminal, the natural comparison is Cursor.
Testing: this sheet was written by a Claude Code session actually running on 19/07/2026, using the features described (file reading, commands, permissions).