Anthropic has launched Cowork, a Claude desktop agent that lets non-technical users hand off real work the way developers use Claude Code, by granting the AI access to a folder on their computer to read, edit, and create files. The feature arrives as a research preview for Claude Max subscribers, priced between $100 and $200 per month, through the macOS app.
The launch is a meaningful step in the race to bring practical AI agents to mainstream users, positioning Anthropic to compete with Microsoft's Copilot in AI-powered productivity rather than conversational chat alone. It also deepens the company's footprint across the consumer AI tools landscape.
Born from unexpected use cases
Cowork grew directly out of how people were already bending Claude Code to non-coding tasks. Anthropic engineer Boris Cherny noted that developers used the coding tool for vacation research, building slide decks, cleaning up email, cancelling subscriptions, and even recovering photos from a hard drive. Recognizing the pattern, the team stripped away the command-line complexity and wrapped the same underlying agent in a more approachable interface aimed at people who never touch a terminal.
How the folder-based architecture works
Unlike a standard chat where you paste in text, Cowork asks you to designate a specific folder the agent can operate within. Inside that sandbox it can read existing files, modify them, or create new ones. Anthropic offers concrete examples:
- Reorganizing a cluttered downloads folder by sorting and renaming files
- Generating an expense spreadsheet from a pile of receipt screenshots
- Drafting a report from notes scattered across multiple documents
The system runs on an agentic loop: it forms a plan, executes steps, checks its own work, and pauses to ask for clarification when it gets stuck. Built on the Claude Agent SDK, it shares the same architecture as Claude Code and can pair with connectors like Asana and Notion, plus the Claude in Chrome extension for web tasks.
A recursive build story
One striking detail is the speed of development. Anthropic employees said the team built Cowork in roughly a week and a half, largely using Claude Code itself. That has prompted industry observers to point to a recursive improvement loop, where AI tools accelerate the building of the next generation of AI tools.
Why Anthropic is warning users
Moving from suggesting edits to actually making them introduces real risk, and Anthropic devoted unusual space in its announcement to caution. The company acknowledged that Claude can take destructive actions, such as deleting local files, if instructed to, and urged users to give very clear, specific guidance. It also flagged prompt injection attacks, where hidden instructions embedded in online content could nudge the agent toward harmful actions, noting that securing real-world agent behavior remains an active area of research.
Availability and what's next
Cowork is currently limited to Claude Max subscribers on macOS, with users on other tiers able to join a waitlist. Anthropic has signaled plans to add cross-device sync and bring the feature to Windows as it learns from the preview, which Cherny described as early and raw, much like Claude Code felt at its own debut. Follow rollout updates in our tech news.
Frequently asked questions
What is Anthropic Cowork?
Cowork is a desktop AI agent that operates inside a folder you choose, reading, editing, and creating files on your behalf. It brings the autonomous power of Claude Code to non-technical users through the macOS app.
Who can use Cowork right now?
It is a research preview limited to Claude Max subscribers on macOS, with plans priced from $100 to $200 per month. Users on other tiers can join a waitlist, and Windows support is planned.
Is it safe to give an AI agent access to my files?
Anthropic warns it carries risk. Because Cowork can modify and delete files, you should scope it to a specific folder, give precise instructions, and be aware of prompt injection threats from untrusted online content.
