What Is Claude Code? How It Differs From Claude Chat

Claude Code is Anthropic’s AI coding agent. It can read a project, propose and apply file changes, and run commands using the tools and permissions available in its session. Instead of copying code between a chatbot and an editor, you can ask it to work within the project itself.

The terminal is one way to use it, but Claude Code also has editor, desktop, and browser interfaces. It is a product built around Claude models, rather than a separate model that runs entirely on your computer.

What Claude Code can do

Typical tasks include explaining unfamiliar code, fixing a bug across several files, writing tests, and reviewing a proposed change. Its access to files helps it follow dependencies that would be cumbersome to paste into a conversation.

A useful first request is narrow: “Explain how this project starts, identify its test command, and do not change files.” Follow that with one small, reviewable change. A working-looking answer is not evidence that the code works; inspect the diff and run the relevant checks.

Claude chat, Projects, and Claude Code compared

The desktop app can also provide a Claude Code interface, so “desktop” and “Code” are not mutually exclusive. Available features depend on the interface, account, and configuration. Consult Anthropic’s overview for supported setup options.

How skills fit in

Skills package reusable instructions in a SKILL.md file, with optional supporting material. Claude can select a relevant skill, or you can invoke one with a slash command. A skill is guidance for performing a task; it does not by itself guarantee correct output or give unrestricted access to files.

For example, a competitor-analysis skill could define the comparison criteria and require sources for pricing claims. Whether the agent can retrieve information or update a report depends on its tools and permissions. See how Claude Skills work for related context.

Choose a setup and check its boundaries

Start with the Windows installation guide or Claude Code in VS Code, depending on how you prefer to work.

Keep a recoverable copy of important files and review the session’s permission mode. Approval behavior varies by mode and organizational settings. Local project files do not mean that all processing stays offline: distinguish where files are stored from where the AI service processes requests. Check Anthropic’s security guidance before granting broad access.