Skip to content
Neural Daily

ToolsNews

PostgreSQL 19 delayed as AI code reviews find bugs

PostgreSQL 19 was due in September and is now expected on November 19. Percona developer Zsolt Parragi says AI-driven code reviews found many of the issues behind the reverted features, while the number of security fixes also surged.

By: Mikola Máté Patrik · · 2 min read

Percona Community blog post by Zsolt Parragi: Apologies for delaying Postgres 19, dated 15 October 2026
Image: Percona Community · Screenshot · Screenshot of official page · Source

What happened

In an October 15 blog post, Parragi writes that the pgsql-hackers threads make it clear many reported issues were found by AI reviews (Percona Community). Several major features that had already been committed were reverted and the release slipped more than once. He himself reported issues with FOR PORTION OF, REPACK and online checksums.

Why now

The author points to big jumps in model capability: Fable 5 in June, Opus 5 and GPT-5.6 in July, Opus 5.5 and GPT-6 in September. All arrived after the PostgreSQL 19 feature freeze, so code that had already been reviewed and merged got a much deeper look from the new models. Early in the year AI mostly found minor issues or false positives; now, he says, it finds as many issues in a patch as humans do, or more.

The “vulnpocalypse”

The second factor is a flood of security fixes. According to the post, PostgreSQL shipped between 2 and 8 security fixes a year from 2020 to 2025, and 44 so far in 2026. Of those, 28 landed in August, during the beta. Previously one or two committers handled security fixes for a minor release; in August it was ten, which took time away from development.

What may change

Proposals on the list include minor releases every 1.5 to 2 months and an easier path to becoming a committer. Parragi says design-level review remains AI's weak spot: whether a feature makes sense for PostgreSQL at all still needs human judgement. Some reverted features may return in PostgreSQL 20.

Why it matters

This is one of the first well-documented cases of AI review visibly reshaping the release schedule of a major open-source project. The lesson for teams: re-review code already on main with newer models from time to time, and expect a spike in findings whenever a stronger model ships.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles