In short
- Linus Torvalds released Linux 7.4-rc1 on 1 Nov 2026: almost 20k commits instead of the usual 15k, many of them fixing issues AI found.
- Robert Haas writes that PostgreSQL 19 will most likely slip to November, because AI finds more bugs and humans still have to review and fix them.
In short: two projects that a large part of the internet runs on now show in numbers what happens when AI starts hunting for bugs at scale. It finds more of them than humans can calmly review.
Linux: the biggest change window in a while
- On 1 Nov 2026 Linus Torvalds released Linux 7.4-rc1 and called this merge window big: almost 20k commits instead of the usual ~15k.
- He writes that many changes fix issues found by AI, and AI increasingly suggests the patches too, although maintainers often rewrite them.
- He is happier that this wave arrived now rather than as fixes during the stabilisation phase, and hopes AI work is becoming a normal part of development.
- He notes that a full third of the patch is AMD graphics register definitions, and that people expect 7.4 to be the next LTS release, which also explains the size.
PostgreSQL 19: late because of too many findings
- Robert Haas, a long-time PostgreSQL committer, writes that version 19 will most likely not ship until November, instead of the usual September or October.
- From the feature freeze (8 April) to 25 September, 1,169 commits went to the version 19 branch, against 724 to the version 18 branch a year earlier.
- Fixes to back branches (the previous release) rose from 248 to 433, more than 74%. He links this to people reporting bugs that AI found.
- Between 27 August and 15 September three major features were reverted, partly because reviewers were busy with security issues and older releases.
How to read this
Haas stresses that committers did not get more careless. More of their mistakes are simply found quickly. Fixes drafted by AI are not always correct or up to project standards, so a human still has to read and commit them. The Linux numbers come from Torvalds himself, while the AI effect on PostgreSQL is Haas’s own judgement, not a separate study. Commit counts are a rough signal, not a measure of quality. I write about keeping AI-assisted code safe in the AI code security hub.