What Claude Code actually is, in plain terms

Not an editor, not autocomplete. A command-line agent that reads your repository, runs commands and iterates until a task is done.

Most descriptions make it sound like autocomplete with extra steps. It is closer to the opposite: you describe an outcome, and it works in your repository until the outcome exists.

The loop

You give it a task in a terminal. It explores the files it thinks are relevant, proposes a change, applies it, runs whatever command verifies it (tests, a build, a type check), reads the output and adjusts. You can stop it at any point, and you approve the plan before it starts if you ask it to plan first.

That last clause is the whole skill. Asking for a plan, reading it, and only then letting it run moves the expensive exploration to a single pass instead of twenty. The same task, unreviewed, costs more and usually lands somewhere slightly wrong.

What makes it different from an editor assistant

• It acts, not just suggests. It edits files and runs commands rather than producing a diff you paste somewhere.
• It works without you. A run can last minutes; you are not steering every step.
• It reads context on demand. Instead of you pasting files, it opens what it needs — which is also why it burns quota faster than chat.
• It is the same assistant in every shell. Nothing about it lives in a sidebar, so there is no editor to keep open while it works.

What it is bad at

Broad, ambiguous requests. "Make the app better" produces expensive wandering. Narrow, verifiable tasks — "make this failing test pass", "add pagination to this endpoint", "split this nine-hundred-line file" — are where it earns its cost. If you cannot write the stop condition, the run will not have one, and it will keep going until it hits a limit.

What you need to run it

Node.js, a terminal and a plan that includes Claude Code. Pro includes it with modest limits; the Max tiers exist because the limits are the thing you run into, not the features. There is no separate purchase for the tool itself — it comes with the subscription, and the subscription is what the money goes to.

How the quota disappears

Two windows govern it: a rolling block of roughly five hours, and a weekly ceiling underneath it. A long agent loop re-reads context every turn, so bursts are what consume the short window; steady all-day usage is what meets the weekly one. Continuing a healthy session is usually cheaper than starting a new one, because the new one has to rebuild context from scratch. The practical version of that advice is short: one session per task, a scratch file for the decisions made, and stop opening a fourth tab "for the other approach".

How people actually use it

Usually alongside an editor that handles small edits better. The common split is: editor for renames and type fixes, agent for anything measured in "implement this feature". Running both against the same file at the same time is the one combination to avoid — the conflict comes out on your side of the screen.

What it costs here

Claude Pro is 105 CNY for one month, about 14 euro, with the official equivalent of 134 CNY printed next to it. Max 5x is 523 CNY, about 70 euro, and Max 20x is 1047 CNY, about 140 euro. Every plan is one month and does not renew. Activation needs one thing from your side: the Organization ID from Settings, then Account, starting with org_ — never a password, a 2FA code or a key starting with sk-ant-. Payment is crypto only, and no email is sent at any stage.

If you are sizing the plan, the rate-limits guide explains where the ceilings are, the install page covers getting it running, and the plan list shows what each tier costs here.

Common questions

Back to plans →