Linus, the visionary behind Linux, has openly adopted Claude Code for the development of kernel functionalities. In the recently introduced branch, aptly named 'llm', he tackles the cache line thrashing bottleneck—a challenge arising from frequent modifications to atomic reference counts in the original multi-threaded concurrent scenarios involving small file reads. By leveraging the Read-Copy-Update (RCU) mechanism, he eliminates the necessity for reference count adjustments on the read path, leading to a remarkable throughput surge of approximately three times in an 8-thread test environment.

Furthermore, Linus harnessed the power of large models to meticulously review old code. This process involved not only verifying reported bugs but also unearthing three latent defects within the data structure library—flaws that had eluded detection for over a decade. The pragmatic Linus, who once dismissed 90% of AI as mere marketing hype, has now, through his actions, underscored the tangible benefits of AI-assisted development. His stance emphasizes that having a well-defined development direction and the ability to discern code correctness outweigh the mere act of manually typing out code.