Robert Haas: LLM-found bugs are why PostgreSQL 19 is late

PostgreSQL major releases have landed between mid-September and mid-October for about a decade. PostgreSQL 19 will not. Robert Haas, a Postgres committer who works at Databricks, gives his explanation in a Neon blog post dated 21 October: “LLMs are finding bugs faster than committers can fix them.”

Haas writes that PostgreSQL 19 “most likely won’t be released until November, and it’s not impossible that the project could decide to postpone the release further.” The release team’s current plan, which we covered on 11 October, is an RC2 on 12 November and general availability on 19 November. The project roadmap page still said “planned for October 2026” when we read it on 23 October.

The numbers he uses

Haas compares the window from 8 April (feature freeze) to 25 September in 2025 and 2026:

• master: 1,037 commits in 2025, 1,443 in 2026.
• The release branch: 724 commits to REL_18_STABLE in 2025, 1,169 to REL_19_STABLE in 2026.
• The previous release: 248 commits to REL_17_STABLE in 2025, 433 to REL_18_STABLE in 2026.

From those counts he concludes that commits made to master only, without a back-patch, “actually declined slightly” from last year, while back-patched commits “jumped by more than 60%.” By his count, commits back-patched to at least one older release as well rose by more than 74%.

His reading: committers were about as careful as before, but “a huge number of people on the Internet are now reporting bugs in PostgreSQL that they found using LLMs.” He says LLMs can draft fixes too, but those fixes “aren’t always correct or compliant with PostgreSQL standards,” so humans still review and commit them.

Why that delays 19

Haas names two effects. Fixing more bugs takes more committer time. And time that would have gone into reviewing what was headed for 19 went into security issues and back-branch bugs instead, so it took longer to decide what to pull.

That is his explanation for three late reverts: UPDATE/DELETE FOR PORTION OF, SQL/PGQ property graph queries, and ALTER TABLE MERGE/SPLIT PARTITION, all removed between 27 August and 15 September. Our earlier report covers those reverts.

What he expects next

Haas does not think a short delay is worrying, since it “emerges directly from the project’s concern for quality.” The longer view worries him more: with “forty years of bugs” that LLMs can now find, he asks “how are we ever going to catch up, and how will PostgreSQL continue to innovate?” He also writes that every feature from now on “will be scrutinized by AI.”

This is one committer’s analysis on his employer’s blog, not a statement from the PostgreSQL release team. The counts are his, and he says himself that more committer mistakes could explain part of the rise.