Postgres 19 Delayed to November as AI Bug Reports Flood In

Why is Postgres 19 late?

The short answer from the project itself: LLMs. According to PostgreSQL developer Robert Haas, the release slipped because AI tools are finding bugs faster than humans can review and fix them.

PostgreSQL major releases have followed a reliable rhythm for a decade: no earlier than mid-September, no later than mid-October. This year, PostgreSQL 19 most likely won't ship until November, and Haas doesn't rule out a further delay.

The numbers back him up. Between April 8 and September 25, commits back-patched to the stable branch jumped from 724 in 2025 to 1,169 in 2026 — a rise of more than 60%. Back-patches to at least one prior release climbed even faster, from 248 to 433, a 74% increase.

Meanwhile, commits to the master branch actually declined slightly. The project isn't writing more code; it's fixing more of it.

Late reverts: the features that didn't make it

Three sizable features were pulled between August 27 and September 15, well after the April 8 feature freeze.

UPDATE/DELETE FOR PORTION OF, SQL Property Graph Queries (SQL/PGQ), and ALTER TABLE … MERGE/SPLIT PARTITION all got the axe late in the beta cycle. Major reverts right after feature freeze aren't unusual; multiple reverts this late in the game are.

Haas points to two causes. First, more bugs means more fixes, and LLM-drafted patches still need human review and must meet PostgreSQL's standards. Second, time that would have gone into reviewing release content got diverted to security issues and back-branch bugs, delaying decisions about what to revert.

The long-term worry: forty years of bugs

The delay itself isn't alarming, Haas writes. The bigger question is how the project catches up.

PostgreSQL's git history reaches back more than 30 years, with code written a decade before that. That's a lot of latent bugs that humans never found — and LLMs are now finding them.

Haas is candid: he isn't sure how the project catches up or keeps innovating at the same pace. He also defends the codebase, calling it some of the cleanest he's worked on, and notes that if it had more human-visible defects, AI would likely find even more.

His conclusion is blunt: every feature written from now on will be scrutinized by AI, and things that once went unnoticed for years will surface in hours. That's a challenge and an opportunity.

What it means for you

For teams running PostgreSQL in the US, UK, or EU, the practical impact is timing. If you were planning an upgrade around the usual October release, budget for November instead, and watch the release notes for the reverted features — SQL/PGQ in particular was expected to be a headline item.

Managed Postgres providers will follow the upstream release, but their upgrade windows vary. Check with your provider before scheduling any migration.

Postgres 19 will arrive when it's ready, not when the calendar says so. The delay is a quality signal, but the surge in AI-found bugs is a warning for every large codebase: finding defects is getting cheap, and fixing them is not.