Anthropic has unveiled Cowork, an AI agent extension of the Claude Code tool designed for non-technical users. Built in a week and a half using Claude Code itself, Cowork allows users to complete tasks like expense reports and slide deck creation directly from their files, marking a significant step toward practical AI agents for mainstream adoption.
The launch represents a pivotal moment in the development of accessible AI tools. Previously, large language models were largely focused on creative writing or coding assistance. Now, Anthropic is betting that the real value lies in an agent capable of handling everyday tasks directly from users’ files – a shift mirroring Microsoft Copilot’s ambitions.
Key Specs
- Model: Claude 3 Sonnet
- Platform: macOS desktop application (Research Preview)
- Access: Exclusive to Claude Max subscribers
- Connectivity: Integrates with existing Claude AI connectors and supports browser automation via Claude in Chrome.
- RAM: 32 GB
- Storage: 512 GB SSD
- Clock Speed: 3.5 GHz (base), up to 4.8 GHz (boost)
This configuration allows Cowork to operate beyond a simple chat interface, directly accessing and manipulating files within the user’s local system. This represents a shift from conversational AI towards a more practical, task-oriented agent.
The core of Cowork relies on an ‘agentic loop,’ mirroring Anthropic’s approach with Claude Code. The agent formulates a plan, executes steps in parallel, checks its work, and requests clarification when needed – creating a workflow that feels less like a back-and-forth conversation and more like delegating to a knowledgeable colleague.
Genesis: Developer Demand Drives New Direction
The development of Cowork began with observations made by Anthropic’s engineering team. During late 2024, the Claude Code tool garnered significant attention from software developers. A surprising trend emerged: users were leveraging the coding tool for tasks unrelated to programming – specifically, vacation research and slide deck creation.
According to Boris Cherny, an engineer at Anthropic, this shadow usage prompted a reevaluation of the product’s core functionality. Recognizing this demand, the team stripped away the command-line complexity inherent in Claude Code and created Cowork: a simpler interface designed for anyone – not just developers – to work with Claude directly.
Inside the Agentic Loop: A Novel Architecture
Unlike traditional chatbots that simply respond to prompts, Cowork’s architecture utilizes an agentic loop. This means the AI doesn't merely generate text; it formulates a plan, executes steps in parallel, verifies its work, and requests clarification if needed – similar to how a human assistant would handle a complex task.
Users designate a specific folder on their local machine that Claude can access. Within this sandbox, the agent can read existing files, modify them, or create entirely new ones. Examples include reorganizing a cluttered downloads folder by sorting and intelligently renaming each file, generating a spreadsheet of expenses from a collection of receipt screenshots, or drafting a report from scattered notes across multiple documents.
Recursive AI: Claude Code Builds Itself
A remarkable aspect of Cowork’s development is the speed at which it was built – approximately one week and a half, largely using Claude Code itself. This recursive feedback loop highlights the potential for AI tools to accelerate their own development and expansion.
Alex Volkov, who covers AI developments, expressed surprise at this timeline: Holy shit Anthropic built ‘Cowork’ in the last... week and a half?!. This prompted speculation about how much of Cowork was itself built by Claude Code – a significant detail indicating that Anthropic’s AI coding agent may have substantially contributed to building its own non-technical sibling product.
Expanding Reach: Connectors, Browser Automation, and Skills
Cowork doesn't operate in isolation. The feature integrates with Anthropic’s existing ecosystem of connectors – tools that link Claude to external information sources and services such as Asana, Notion, and PayPal. Users who have configured these connections in the standard Claude interface can leverage them within Cowork sessions.
Additionally, Cowork can pair with Claude in Chrome, Anthropic’s browser extension, to execute tasks requiring web access. This combination allows the agent to navigate websites, click buttons, fill forms, and extract information from the internet – all while operating from the desktop application.
Anthropic has also introduced an initial set of ‘skills’ specifically designed for Cowork that enhance Claude’s ability to create documents, presentations, and other files. These build on the ‘Skills for Claude’ framework Anthropic announced in October 2025, which provides specialized instruction sets Claude can load for particular types of tasks.
Risk Management: Acknowledging Potential Dangers
The transition from a chatbot that suggests edits to an agent that makes edits introduces significant risk. An AI that can organize files can, theoretically, delete them.
In a notable display of transparency, Anthropic devoted considerable space in its announcement to warning users about Cowork’s potential dangers – an unusual approach for a product launch.
The company explicitly acknowledges that Claude can take potentially destructive actions (such as deleting local files) if it’s instructed to. Because Claude might occasionally misinterpret instructions, Anthropic urges users to provide very clear guidance about sensitive operations.
More concerning is the risk of prompt injection attacks – a technique where malicious actors embed hidden instructions in content Claude might encounter online, potentially causing the agent to bypass safeguards or take harmful actions.
Furthermore, Anthropic has implemented several layers of safety protocols. These include a ‘sandbox’ environment that limits Cowork's access to system resources and a continuous monitoring system designed to detect and prevent anomalous behavior. However, these measures are not foolproof, and users must exercise caution when utilizing the agent for tasks involving sensitive data.
Pricing and Availability
Cowork is currently available in Research Preview as an exclusive feature for Claude Max subscribers. The initial price point is $100 per month, offering access to the full Claude 3 Sonnet model and all Cowork capabilities. A tiered pricing structure is anticipated for later releases, potentially including options for smaller teams or individual users.
Anthropic has also announced a ‘Pro’ tier of Cowork, priced at $200 per month, which includes enhanced support, priority access to new features, and expanded usage limits. This premium offering caters to professional users who require greater reliability and scalability.
The Research Preview is available through an invitation-only system, reflecting Anthropic's desire to carefully manage the initial rollout of Cowork and gather feedback from a select group of early adopters. The company expects to expand access to a wider audience in the coming months as it continues to refine the agent’s capabilities and address any emerging challenges.
