Incident sheet · /infrastructure
- System
- Infrastructure & Cloud
- Filed
- Read
- 4 min · 733 words
- Severity
- Routine
PostgreSQL 19 shipped late as AI-found bugs strained maintainers
PostgreSQL 19 arrived more than a month outside its normal release window as AI-discovered defects sent maintainers back to older branches, forcing feature deferrals and extra release work.

Event log
- PostgreSQL 19 reached general availability more than one month after its normal mid-October release-window close.
- A normal PostgreSQL major-release cycle takes 14 to 15 months from first commits to general availability.
- PostgreSQL fixed 5 CVEs in February 2026, 11 in May and 28 in August.
- REL_19_STABLE received 1,169 commits from April 8 to September 25, 2026, while master received 1,443.
- Back-patch commits rose more than 60% on the active stable branch and more than 74% across the active and older branches.
By Tom Whitfield
PostgreSQL 19 shipped more than one month after its normal window closed, as maintainers shifted attention from new features to defects AI systems were finding in older stable branches. PostgreSQL contributors Jonathan Katz and Robert Haas said: "The delay came directly from the project's concern for quality." Quality won; several features did not.
The version reached general availability after a process that normally takes 14 to 15 months from first commits. PostgreSQL divides that time into two phases and keeps a fixed annual cadence: no major release before mid-September and none after mid-October. For most of the past decade, the schedule held.
The backlog reached the back branches
The project began as POSTGRES at UC Berkeley in the 1980s and has developed in the open since 1996. No single company owns it. A worldwide developer community maintains it around one major release each year, with reliability and predictability guiding the calendar.
That norm matters because PostgreSQL's codebase is large, complex and old, with a huge number of bugs. Git history, converted from CVS, extends more than 30 years; its first code predates that history by another decade. Features that cannot stabilize on schedule normally move to a later major version, preserving the release date.
Around February 2026, Katz and Haas say, the latest AI models became capable enough to write real production code, which they describe as less buggy, while also helping developers move faster. Code review and security research demanded less human expertise, and the volume of issues LLMs could identify rose sharply. AI also made fixes easier, though not to the same degree as discovery. Finding accelerated faster than repair; committers still did the work.
- From August 2023 through November 2025, each quarterly update fixed one to four CVEs.
- In 2026, PostgreSQL fixed five CVEs in February, 11 in May and 28 in August.
- Quarterly bug fixes rose from roughly 50 to 75 per release to about 140 in August 2026.
Not every CVE carries the same severity; many fixes address long-tail issues. Even so, the project's normal update volume had changed materially.
The commit record shows the same pressure. Between April 8 and September 25, 2025, PostgreSQL master received 1,037 commits and REL_18_STABLE received 724. During the same date range in 2026, master received 1,443 commits and REL_19_STABLE received 1,169.
Within the narrower period from feature freeze to September 25, master commits declined slightly year over year. Back-patched commits to the active stable branch jumped by more than 60%. The authors say contributors probably remained as conservative in 2026; AI systems simply found more of their mistakes faster.
Older branches absorbed an even sharper increase. In the equivalent 2025 period, PostgreSQL recorded 248 commits to REL_17_STABLE. In 2026, it recorded 433 commits to REL_18_STABLE, a rise of more than 74% for the release in progress and at least one older branch.
A huge number of people online began reporting PostgreSQL bugs found with LLMs. Committers then spent substantial time fixing security issues and other defects in maintenance branches rather than reviewing work intended for PostgreSQL 19. A few features in progress were not ready and moved to future releases.
The November 12 update for supported branches contained enough fixes that the release team issued a second release candidate, allowing PostgreSQL 19 to ship with those corrections. The result was a release several weeks later than usual, outside the project's September-to-October window.
What does this change for PostgreSQL users?
Katz and Haas set out the new operating expectation: "Every feature we write from now on will be scrutinized by AI, and bugs that would have gone unnoticed forever in an earlier era will now be found, sometimes in a matter of hours."
For PostgreSQL users, the immediate effect is a delayed release and several features pushed into future major versions. The larger effect sits upstream: contributors now carry a heavier back-patch workload, while AI-assisted scrutiny makes hidden defects harder to leave unexamined. The trade is schedule pressure for better code.
The contributors said the outcome remains uncertain for open-source development. They expect developers to raise their standards, describing AI as both a challenge and an opportunity to make PostgreSQL stronger. The near-term ledger is a delayed release, deferred features and a larger back-patch burden, with more defects caught and fixed.
via databricks.com (Source)
More from Tom Whitfield
Show full bio
Market editor covering people, launches and day-to-day practice in enterprise information technology and the leaders who run at CIO Monitor.
195 articles
The newsroom digest
One email a day with the stories that matter. No noise.
Cross-references
- PostgreSQL 19 Ships Day-One on Lakebase With REPACK and Plan Advice
- Microsoft puts HorizonDB Postgres into general availability at Ignite
- Microsoft previews serverless HorizonDB for agent applications
- Google and Anthropic Compress Vulnerability Disclosure Windows
- OpenAI Acquires Pierre Computer to Strengthen Codex Git Layer