Why Anthropic Thinks AI Should Have Its Own Computer — Felix Rieseberg of Claude Cowork & Claude Code Desktop - Latent Space: The AI Engineer Podcast Recap

Podcast: Latent Space: The AI Engineer Podcast

Published: 2026-03-17

Guests: Felix Rieseberg

What Happened

Anthropic has taken a significant step in AI development by creating Cloud Cowork, a user-friendly version of Cloud Code, designed for non-technical users or those involved in non-coding tasks. In just 10 days, developers leveraged existing technologies to create this platform, which runs on a virtual machine to provide security and autonomy for Claude, the AI, to install necessary software like Python or Node.js.

The decision to run Cloud Cowork locally rather than in the cloud emphasizes ease of access to local files and tools. This approach allows for smoother integration with the user's environment, increasing the platform's effectiveness. Cloud Cowork is evaluated based on typical knowledge work tasks, while Cloud Code remains optimized for coding tasks.

Anthropic's prototype and demo-first culture means they often build multiple versions for user testing. This iterative approach has led to the development of a virtual machine concept, enhancing Claude's capabilities by essentially giving it its own computer. This method of using virtual machines helps separate Claude's operations from the user's computer, enhancing security while maintaining functionality.

Cloud Core, a related product, has evolved from manual task completion to the automation of repetitive tasks through 'skills'. These skills are file-based, easily created, and shared via GitHub repositories, allowing for hyper-personalization and portability. Users have successfully automated tasks like checking calendar conflicts and organizing files, showcasing Cloud Core's versatility.

Anthropic is exploring the broader impact of AI tools on the labor market, particularly concerning junior employees. There is a suggestion to use AI to simulate work experiences, accelerating learning for those new to the workforce. The University of Waterloo's co-op program is cited as a model for preparing students for real-world roles.

The podcast delves into the use of frameworks like Electron, which allows developers to build cross-platform apps using web technologies. Despite its counterintuitive nature of including an entire version of Chromium, Electron provides necessary control and reliability that operating system web views do not. This has proven beneficial in various applications, including the Unreal Engine and Slack app.

Anthropic Labs is working on experimental projects in their internal agent lab, where Cloud Code is a part. One such project, PyQ, allows any repository to be managed by a coding agent with a public webhook, enabling task submissions. These innovations reflect Anthropic's commitment to expanding AI's capabilities and exploring new applications.

Key Insights

• Cloud Cowork was developed rapidly in 10 days, using existing prototypes and technology, emphasizing Anthropic's prototype-first culture.
• Running Cloud Cowork on a local machine rather than the cloud enhances security and accessibility to local files and tools.
• Skills in Cloud Core are easily created and shared, allowing users to automate repetitive tasks and hyper-personalize their workflows.
• Anthropic is concerned about AI's impact on the labor market, especially for junior employees, and suggests using AI to simulate work experiences to aid learning.