A developer successfully used the Claude Fable 5 AI model to write a functional, NT-compatible kernel in Rust, which booted in QEMU and passed self-tests in 38 minutes.

Key Points

• The project, ntoskrnl-rs, consists of approximately 5,100 lines of code across 27 files.
• The kernel successfully implements core components including a scheduler, memory manager, and I/O manager.
• Following the initial 38-minute build, the project evolved over eight days to support unmodified Windows drivers and user-mode binaries like cmd.exe.
• The development process utilized two different AI models: Claude Fable 5 for rapid greenfield generation and Claude Opus 4.8 for iterative debugging.
• The kernel serves as a "trusted computing base," meaning it requires rigorous verification to ensure security and stability for production use.

Why it Matters

This project demonstrates that generative AI can now autonomously architect complex, low-level infrastructure software that was previously considered too costly or risky to rewrite by hand. While the ability to generate code has outpaced current verification methods, this capability suggests a future where legacy C-based systems can be systematically replaced with memory-safe alternatives.