Claude Code

What it is

Claude Code is Anthropic's agentic coding tool. It runs in the terminal, and in editors, and works directly in a codebase: it reads files, runs commands, edits code, runs tests and uses git. A developer gives it a task in plain language and approves or limits what it may do, from asking before every command to working on its own within set permissions.

It can be shaped to a project. An instruction file in the repository tells it the project's rules. Subagents handle narrower jobs in their own context. Hooks run the developer's own scripts when the agent acts, and Model Context Protocol servers connect it to outside systems. Several sessions can run at once, each on its own task.

How I use it

• One developer and a fleet of Claude Code agents build and run Rental Pilot on web, mobile and desktop. The build log publishes the real counts every hour.
• Each repository carries its own instruction file: the money rules, the deploy runbook and what is safe to touch in the legacy application.
• Purpose-built agents compare a behaviour across the legacy, web and phone apps, re-verify defects against deployed code, and answer questions against the database with findings instead of row dumps.
• One agent refuses to approve a change on a passing build alone. It renders the screen in a real browser or the iOS simulator and reports what is actually there.
• Claude Code hooks on my machine push each agent action to the build log over an API Gateway WebSocket: the event, the tool name and the product, with no code or prompts.

Claude Code is the right tool when the work is real code in a real repository and someone experienced is reviewing it. I use it because it works in the terminal next to me, under the same deploy scripts and safety checks I use. The skill is still the old one: deciding what the agents are allowed to touch, and checking what comes back before it ships.

Questions