Anthropic launches Cowork, a Claude Desktop agent that works in your files - no coding required

Anthropic released Claude Code in late 2024, a terminal tool for developers to automate programming tasks. The company observed users employing it for non-programming work, such as vacation research, creating slides, cleaning up emails, unsubscribing from newsletters, recovering wedding photos, monitoring plant growth, and controlling ovens. Anthropic engineer Boris Cherny stated on X that these use cases were diverse and unexpected because the underlying Claude Agent is the best agent and Opus 4.5 is the best model.

Recognizing this 'shadow usage,' Anthropic stripped away the complexity of the developer tool to build a consumer-facing interface. The company blog noted that developers 'quickly started using it for almost everything else,' which 'prompted us to build Cowork: a simpler way for anyone—not just developers—to collaborate with Claude in exactly the same way.'

Cowork requires users to designate a local folder for Claude to access. Within that sandbox, the AI agent can read, modify, or create new files. Anthropic explained on X: 'With Cowork, you give Claude access to a folder on your computer. Claude can read, edit, or create files in that folder.' Examples include organizing a messy Downloads folder, generating a spreadsheet from receipt screenshots, or drafting a report from scattered notes.

The architecture is based on an 'agentic loop': after being assigned a task, the AI doesn't just generate a text response but makes a plan, executes steps in parallel, self-checks, and asks for clarification when stuck. Users can queue multiple tasks and have Claude handle them simultaneously; Anthropic describes this workflow as 'more like leaving a message for a colleague than a back-and-forth conversation.' The system is built on the Claude Agent SDK, sharing the underlying architecture with Claude Code.

The speed of Cowork's development is striking. During a livestream hosted by Dan Shipper, Anthropic employee Felix Rieseberg confirmed that the team built it in about a week and a half. AI observer Alex Volkov expressed surprise on X: 'Oh my god, Anthropic built “Cowork” in… a week and a half?!' This sparked speculation about the extent of Claude Code's involvement. Simon Smith, EVP of Generative AI at Klick Health, stated bluntly: 'Claude Code wrote all of Claude Cowork. Can we agree that we are at least somewhat in a recursive improvement loop?'

If true, this would be one of the most significant examples of an AI system accelerating its own development, potentially widening the gap between AI labs that successfully deploy their own agents internally and those that do not.

Cowork integrates Anthropic's existing connector ecosystem, linking to external information sources and services such as Asana, Notion, and PayPal. Users who have already configured these connectors can leverage them in Cowork sessions. Additionally, Cowork can pair with the Claude in Chrome browser extension to perform tasks requiring web access, such as navigating websites, clicking buttons, filling out forms, and extracting information.

Cherny emphasized that Cowork includes 'many novel UX and safety features,' including a built-in virtual machine for isolation, out-of-the-box browser automation support, support for all claude.ai data connectors, and the ability to ask for clarification when uncertain. Anthropic also introduced an initial set of 'skills' that enhance Claude's ability to create documents, presentations, and other files, based on the Skills for Claude framework announced in October.

The shift from a chatbot that suggests edits to an agent that actually performs them carries significant risks. In its announcement, Anthropic specifically warned users about Cowork's potential dangers, explicitly acknowledging that Claude 'may take destructive actions (like deleting local files) if instructed to do so.' Because Claude may occasionally misinterpret instructions, Anthropic urges users to provide 'very clear guidance' for sensitive operations.

More concerning is the risk of prompt injection attacks—where malicious actors embed hidden instructions in content that Claude reads. Anthropic's transparency is unusual for a product launch and reflects the importance placed on agent safety.