Write a kernel performance patch series with an agent under expert direction

Linus Torvalds published a branch in which Claude Code, under his direction, wrote a Linux patch series that lets many threads read small parts of the same file from the page cache without contending on reference counts, along with the test programs, a review of edge cases and the first draft of the cover letter. He ran the benchmarks on his own machine and disclosed the AI involvement in the commit message; Phoronix reproduced his chart showing a large speed-up for that workload. The lesson is the process: expert review, own measurements and explicit disclosure, which is what the kernel community expects for AI-assisted patches.