Linux 7.2 landed on August 17 with 16,418 non-merge commits and nearly 600,000 new lines of code, making it the second-busiest kernel release in history. Unlike the 6.7 cycle, which was inflated by the bcachefs merge, 7.2's volume is broad-based — spread across drivers, networking, architecture, and tooling. The developer pool hit a new record of 2,652 contributors, continuing a decades-long upward trend whose slope has visibly steepened. The most conspicuous shift is the LLM footprint. Tagged Assisted-by commits jumped from 301 in 7.1 to 1,111 in 7.2 — a 269% increase — and the true count is almost certainly far higher. Many developers omit the tags out of ignorance or indifference, maintainers rarely ask, and at least one maintainer actively strips Assisted-by tags when applying patches. The community's tracking mechanism is broken before it was ever really tested. The effects are already visible in the data. Bug-fix commits carrying Fixes tags hit 4,830, a dramatic spike after years of roughly flat activity. The article attributes this directly to LLMs being pointed at existing code to find bugs — a genuinely productive use case, but one with an expiration date. The stock of existing bugs is finite; the question is whether LLM-generated patches are introducing new ones at the same time they fix old ones. Rosen Penev contributed 146 commits of LLM-authored bug fixes across architecture, driver, networking, and crypto subsystems. First-timer Bryam Vargas submitted 43 commits touching input, NVMe, F2FS, AppArmor, SELinux, NTFS, WiFi, ATA, SCSI, networking, Ceph, and Thunderbolt — a breadth that would challenge a seasoned kernel developer. Neither pattern is inherently alarming, but both raise the question of whether anyone fully understands all the code being merged. Review-tag rates are declining: 48% of commits carried Reviewed-by tags, continuing a mild downward trend. Whether that reflects less actual review or just less tagging discipline is unclear, but the timing is poor. The moment when comprehension of submitted patches is most uncertain is exactly the moment when review coverage should be increasing, not drifting. The employer landscape shows Intel (9.2%), Google (7.3%), Red Hat (5.3%), AMD (5.0%), and Qualcomm (4.7%) as the top corporate contributors by changeset. But the largest single category is "Unknown" at 16.7%, partly reflecting the surge of new developers whose affiliations haven't been mapped. The 613 first-time contributors set a third consecutive record, and the community's ability to absorb, mentor, and quality-check that influx is the binding constraint. The kernel community has absorbed new tools and processes many times over its 35-year history — from BitKeeper to Git, from email patches to structured review workflows. But LLMs represent something qualitatively different: they enable contribution without comprehension. The experiment now running on the kernel's development process is whether a codebase can remain maintainable when a growing fraction of its authors may not fully understand what they've written. The answer will take years to arrive, and the kernel — used in production everywhere from phones to satellites — is the test subject.
