Anthropic Claude Code

Claude Code is Anthropic's agentic coding system, a tool that takes a development task described in plain language and carries it out across a real codebase. It runs in the terminal, in a desktop app, and inside IDEs, and rather than suggesting the next line as you type, it works at the level of a whole project: it reads the codebase, plans an approach, makes changes across multiple files, runs tests, fixes failures, and commits the result. The developer sets the objective and decides what ships, while the execution loop runs on its own.

Control and safety are built into how it operates. By default Claude Code is cautious, asking for permission before it modifies files or runs commands, and it offers permission modes, checkpoints, sandboxing, and a plan mode where you review its proposed steps before it acts. An automatic mode can route each action through a safety classifier to separate routine work from riskier operations. Because it works inside the developer's own environment using their existing tools, nothing happens in an opaque backend.

The tool is designed to be extended. A CLAUDE.md file gives it standing conventions and commands for a repository, and developers can add skills, subagents, slash commands, and hooks, then bundle them as plugins from a growing directory. Through the Model Context Protocol, Claude Code reaches beyond the codebase to databases, internal APIs, issue trackers, and monitoring systems, so it can, for example, read a ticket's acceptance criteria or inspect a production database before writing a migration. The same agentic loop is available programmatically through the Claude Agent SDK for building custom agents, and multiple Claude Code instances can work in parallel on different parts of a large task under a coordinating lead.

Claude Code is powered by Anthropic's Claude models and reflects the company's research on agent safety and human oversight. While built for engineers handling large refactors, debugging, and code review, it has also become an entry point for founders, product managers, and operations teams who can describe an outcome and get working software back.