AI can now build an OS from scratch, claims Anthropic
What happened
Anthropic announced that its AI model, Claude Fable 5.2, built a functioning operating system from the ground up in eight days. The project used more than 1,000 AI agents working together, starting from an empty folder and producing an OS that could boot on a laptop. In a separate experiment, the same model formalized the proof of Fermat's Last Theorem in 30 hours with minimal human involvement. Both demonstrations were conducted as part of Anthropic's efforts to showcase the model's ability to coordinate large numbers of AI agents on complex tasks.
Key details
- Claude Fable 5.2 coordinated over 1,000 subagents for the OS project.
- The operating system included a core, drivers, file storage, windows, a login screen, and an installer.
- The OS was completed in eight days, running continuously.
- For the Fermat's Last Theorem proof, 1,000 agents completed the task in 30 hours.
- The proof was seven times shorter and used a third of the action tokens compared to a previous attempt.
Why it matters
Software developers and technology companies now face the possibility of automating complex engineering projects that previously required large teams and extended timelines. Anthropic's demonstration suggests that AI models can independently manage and execute sophisticated tasks, potentially reducing the time and cost of software development. Organizations may need to adapt their workflows and staffing as AI takes on a larger role in code creation and problem-solving.