Claude Chat, Code, and Cowork — Same Model, Different Permissions

Claude Chat, Code, and Cowork — Same Model, Different Permissions

People confuse Anthropic's three tools not because the names are similar, but because they genuinely sit on the same model. The difference isn't intelligence. It's what each one is allowed to touch on your machine.

1. Claude Chat — conversation

The browser and mobile interface. It answers, writes, and organizes. It does not go browsing your file system. It works only with what you paste or upload.

2. Claude Code — an agent for codebases

Officially: "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools." Contrary to common belief, it is not terminal-only — it runs in the terminal CLI, VS Code and JetBrains extensions, a desktop app, and the browser, all sharing one engine. It edits files, runs tests, commits, and opens PRs. It's a developer tool because its object is a repository, not because the window is black.

3. Claude Cowork — an agent for work that isn't code

Anthropic describes Cowork as using the same agentic architecture that powers Claude Code, with no terminal required. It handles multi-step tasks while you're in a meeting or away from your desk.

• Platforms: web, desktop (macOS, Windows, Windows ARM64, Linux), mobile (beta), ChromeOS
• Plans: paid plans (Pro, Max, Team, Enterprise)
• Scheduling: tasks can run unattended on a cadence you set

4. The real difference is permission

Anthropic's stated control model for Cowork:

• You pick the folders and tools. In their words: "You choose the folders and tools. Claude can't reach anything else."
• Claude shows its plan and waits for authorization before significant actions.
• Deleting anything requires your approval.
• Enterprise admins configure team permissions, spend limits, and per-department tool access, with usage analytics and OpenTelemetry observability.

Saying "it touches local files, therefore it's dangerous" is a half-truth. Risk comes from permission — and so does control. The question isn't "does the AI touch files," it's "which folders will we open, and who decides that."

5. Summary