Claude Code is Anthropic’s coding agent: a program that runs in your terminal, takes a task in plain English, and works on your actual codebase — reading files, editing them, running commands and tests — until the task is done. That one sentence separates it from everything people confuse it with, so let’s take the confusions in order.
It’s not a chatbot
Claude on the web answers questions about code you paste in. Claude Code operates on the real thing. Ask it to “fix the failing tests” and it runs your test suite, reads the failures, opens the relevant files, edits them, and re-runs until green. You approve its actions — it asks before editing files or executing commands — but you’re reviewing work, not copy-pasting suggestions.
It’s not an editor either
Tools like Cursor and Copilot make you faster while you type. Claude Code doesn’t care whether your editor is even open. It lives in the terminal, which makes it scriptable: it can run interactively when you want a pair programmer, or headlessly (claude -p "…") inside CI, cron jobs, and pipelines when you want an employee. The full comparison is in Cursor vs Claude Code.
What it can actually do
- Explore and explain an unfamiliar codebase
- Multi-file edits: features, refactors, migrations, bug fixes
- Run tests and builds, read the failures, iterate to green
- Git: commit, branch, resolve conflicts, open pull requests
- Reach beyond code via MCP — your database, your docs, your task list
- Be extended with hooks, subagents, and skills that encode your team’s workflows
What it costs
It’s included with a Claude subscription — Pro at $20/month covers light use, Max ($100–200) covers daily agent work — or you can pay per token via the API. Details in the pricing guide.
Why it matters more than “AI autocomplete”
Autocomplete compresses the minutes. An agent removes the task. Once a tool can genuinely finish scoped work — write the tests, do the rename, draft the migration — the interesting question stops being “how good is the model” and becomes “which of my tasks should I hand off, and how do I keep track of what I delegated?” That’s the gap Lume fills: a to-do list where a task can be assigned to Claude Code the way you’d assign it to a teammate — it works, you review, you check it off. Start with installing Claude Code; you’ll be delegating within the week.