Anthropic Launches Cowork, Letting Claude Autonomously Manage Files on Mac

Anthropic has launched Cowork, a new feature that allows its AI assistant Claude to autonomously manage files and execute tasks on a user’s Mac.

The tool, released on January 12, extends Claude beyond text-based interaction into direct file system operations. Users can point Claude at a folder, after which it can complete tasks such as building spreadsheets from receipt screenshots, organising downloads by file type or content, and generating reports from scattered notes without requiring step-by-step prompts.

According to Anthropic, Cowork was developed after observing how users were repurposing Claude Code, the company’s developer-oriented product, for non-technical workflows. Claude Code was originally designed for software development, but users were increasingly using it to carry out personal and workplace tasks, including creating travel plans, recovering corrupted photo libraries, tracking plant growth, and controlling connected appliances through terminal commands.

Anthropic said this pattern indicated broader demand for AI systems capable of executing multi-step tasks independently rather than only providing guidance or recommendations.

What Cowork Actually Does

Give Cowork access to a designated folder, and it operates inside that sandbox. It reads files to understand context, creates new documents when needed, reorganizes existing ones based on content rather than arbitrary naming conventions. The system runs on the same Agent SDK that powers Claude Code, but strips away the terminal complexity.

Early users have deployed it across use cases Anthropic never anticipated. “Vacation research, building slide decks, cleaning up email, cancelling subscriptions, recovering wedding photos from a hard drive, monitoring plant growth, controlling your oven,” according to Anthropic engineer Boris Cherny’s list of actual implementations.

The tool currently works exclusively on macOS for Claude Max subscribers—those paying $100 to $200 monthly for premium access. Other subscription tiers can join a waitlist. Windows support is in development. Cowork also taps into Claude’s existing integrations: Google Drive, Canva, and other connected services become accessible within its automation workflows.

The Productivity Gamble

Here’s the tension: AI companies keep promising productivity breakthroughs through agentic software, but corporate data shows most implementations fail to deliver measurable gains. Workers end up supervising AI instead of being liberated by it.

Anthropic’s counter? Software engineers already trust Claude Code for production work—code that ships to customers, handles real transactions, manages critical infrastructure. They wouldn’t use it if the output required constant clean-up. Cowork uses identical architecture, just aimed at different tasks.

But the company isn’t glossing over risks. Anthropic devoted unusual space in its announcement to warnings about prompt injection attacks and Cowork’s file deletion capabilities. The recommendation: make instructions explicit, back up your designated folder regularly, understand you’re giving an AI system real operational control over your files.

That transparency distinguishes Anthropic’s approach from competitors like OpenAI’s Operator agent (released January 2025) and Microsoft’s Copilot suite, which emphasize seamless integration over user awareness of system boundaries.

Built by AI, For Everyone

According to sources close to the development, Anthropic built Cowork in about a week and a half—using Claude Code itself for most of the engineering work. The recursive development process, where AI tools build better AI tools, suggests how quickly capabilities are compounding.

The speed also reveals competitive pressure. ChatGPT became shorthand for conversational AI barely two years ago. Now the entire industry is pivoting toward agents that act autonomously. The question isn’t whether AI assistants will handle tasks independently—it’s who builds the version people actually trust.

Anthropic is betting the answer lies in giving users control over scope and visibility. You define the sandbox. You approve access. You see what Claude does inside that folder. The alternative—AI that operates seamlessly across your entire system with minimal transparency—might be more convenient but raises obvious concerns about control and accountability.

Whether Cowork becomes standard workflow infrastructure or another overhyped productivity tool depends entirely on execution. Can it handle the messy, context-dependent work that fills most office jobs? Or does it create more work through errors that require human clean-up?

Those answers will come from actual usage, not launch announcements. For now, Cowork moves AI assistants one step closer to operational partners rather than conversational interfaces. The chatbot metaphor is fading. What replaces it is still being defined.