Claude Code and Cowork can now use your computer

The line between a chatbot and an autonomous agent has just been blurred. Anthropic, one of the major players in the language model market, has announced a breakthrough update for its tools Claude Code and Claude Cowork. From now on, the artificial intelligence does not just answer questions but literally takes the reins of the user's computer to perform complex tasks within the system environment. This is a move that positions Anthropic ahead of the competition in the race for the title of the most useful digital assistant.

The new functionality allows Claude models to interact directly with the operating system. The AI has gained the ability to open files, use a web browser, and run development tools (dev tools). This is a significant paradigm shift – instead of copying code from a chat window to an editor, the user can now instruct Claude to independently perform operations within the local work environment. This is the realization of the "computer use" vision that Anthropic has been working on intensively in recent months.

Operational Architecture and Service Prioritization

The mechanism behind the new features is based on an intelligent access hierarchy. First and foremost, Claude AI attempts to use dedicated connectors for supported external services, such as the Google Workspace suite or the Slack messenger. Thanks to this, integration with calendars, documents, or company communication takes place in a structured manner optimized for specific APIs. Anthropic focuses on workflow fluidity, where the chatbot becomes a central hub managing applications.

However, if a dedicated connector is not available, the system does not give up and attempts to perform the assigned task by operating directly on the computer's interface. This flexibility makes Claude Code a powerful tool in the hands of developers, while Claude Cowork – which debuted in January as a more accessible, consumer version of the agent – transforms into a true virtual office assistant. The ability to navigate the operating system opens the door to process automation that previously required manual clicking and switching between windows.

Security in the Era of Autonomous Agents

Giving control over the cursor and file system to artificial intelligence raises natural security concerns. Anthropic is aware of this, which is why it has introduced a permission request mechanism before the AI takes key actions. Nevertheless, the company maintains a high degree of caution, officially recommending that, at this stage, the feature should not be used to handle sensitive information. This is a signal that the technology, while impressive, is still in an intensive testing phase.

• Claude Code: Focused on software engineering, operating on repositories, and testing code.
• Claude Cowork: Designed for non-technical users for daily office and organizational tasks.
• Required Permissions: Every significant system action requires authorization from the user.
• Data Limitations: Recommendation to avoid processing critical data and passwords.

Currently, the computer use feature is available as part of a research preview phase. This means that Anthropic intends to actively collect feedback from the community to refine decision-making algorithms and eliminate potential errors in graphical interface interpretation. This is an approach typical of industry leaders who prefer to release raw but functional technology to a narrow group of recipients rather than waiting for a perfect version.

Availability and the Dispatch Ecosystem

However, the new capabilities are not available to everyone. Anthropic has limited access to subscribers of the Claude Pro and Claude Max plans, clearly indicating the professional positioning of these tools. Furthermore, at this moment, support is exclusively for macOS. Apple computer users are thus becoming the test group for technology that has the potential to redefine how we interact with operating systems altogether.

A key element of this update is also the integration with the Anthropic Dispatch feature. This allows for maintaining a single, continuous conversation with the chatbot across various devices – from a smartphone to a desktop computer. As a result, a user can start giving instructions on their phone, and Claude will finish the task directly on the computer at the office or home, having access to the necessary files and tools. This solution drastically reduces friction in the mobile-to-desktop workflow.

The introduction of full interaction with the operating system by Claude is a turning point for the entire AI industry. Anthropic proves that the future does not belong to chatbots confined to a browser tab, but to system agents who understand the context of our local files and tools. While current limitations to macOS and paid subscriptions may seem like a barrier, they are a necessary step toward the safe autonomy of artificial intelligence. It can be assumed with high probability that the success of this testing phase will force competitors like OpenAI or Microsoft to implement similar solutions on a large scale even faster.