Claude Code

Anthropic's agentic coding tool — plans, edits and ships real changes from the terminal.

Overview

Claude Code is Anthropic's agentic coding tool. It runs in the terminal (and inside editors and the desktop app), reads your repository, plans a change, edits files, runs commands and tests, and can open the pull request — working from a task description rather than line-by-line autocomplete. It is the agent-first model of coding, as opposed to the editor-first model of tools like Cursor.

How I use it

Claude Code is the agent I hand real tasks to. A typical one: "the contact form should require a phone number across the whole stack, without breaking historical rows." It reads the relevant files, proposes a plan, makes the change in the frontend, the edge function and a migration, runs the test suite and opens a pull request I review on GitHub . Most of the changes to this site over the last months shipped that way. I use it alongside Codex rather than instead of it — two agents reading the same repository catch different things, and one reviewing the other's pull request is cheap insurance. Cursor is where I go when I want to stay close to the code myself. The split is described in Build with AI on my stack page.

Key features

Works from a task: plans, edits across files, runs commands and tests, opens PRs Deep codebase understanding — it reads before it writes, and explains what it found Hooks, slash commands, subagents and CLAUDE.md project memory for repeatable workflows Runs in the terminal, VS Code/JetBrains, the desktop app and GitHub Actions Permission model: every file write and command is visible and approvable Included in Claude's paid plans rather than metered separately (API use is an option)

Best for

Builders who can describe a change precisely and review a diff critically. It shines on codebases with tests, because it runs them. It is a poor fit for someone who cannot evaluate the code it produces — the agent is fast, but the judgement is still yours.

Where it fits in my stack

Claude Code → GitHub (every change is a pull request), Supabase (migrations and edge functions), Vercel (preview deployments for every PR), paired with Codex and Cursor . Related reading: my Claude Code review and Claude Code vs Codex .

Pros

Genuinely autonomous on multi-file tasks — plan, change, test, PR Excellent at explaining an unfamiliar codebase before touching it Project memory and hooks make team conventions stick Bundled with Claude Pro/Max/Team, so no separate coding subscription

Limitations

Usage limits on subscription plans — heavy days hit the ceiling Terminal-first: less visual than an IDE for exploring diffs Long tasks need clear scope; vague briefs produce confident detours Runs commands on your machine — permissions deserve attention

Pricing

Claude Code is included in Claude's paid plans. As listed on claude.com/pricing: Pro at $20 per month ($17 per month billed annually), Max from $100 per month, Team at $20–25 per seat per month (Premium seats $100–125), and Enterprise. The Free plan does not include Claude Code. It can also run against API usage billing. See the official pricing page .

Verdict

Claude Code is the tool that moved AI from "helps me type" to "does the task while I review". I keep it because it turns a written scope into a reviewed pull request, and because pairing it with a second agent makes the output safer, not just faster.