Key takeaways
- PostgreSQL 19 now targets general availability around November 19, The Register reports, after a late September or early October plan.
- Between 8 and 10 significant features, including SQL/PGQ graph queries, were pulled and pushed to PostgreSQL 20.
- Committer Peter Eisentraut says AI assisted code review found problems faster than the team could fix them.
The PostgreSQL 19 delay is now official in all but name. The next major version of the popular open source database, once expected in late September or early October, is now aiming for general availability around November 19, according to The Register. Between 8 and 10 significant features have been pulled from the release and pushed to PostgreSQL 20, and one of the reasons may surprise people who think AI only speeds software up: AI code review found more problems than the volunteers could fix in time.
That is not a story of AI writing bad code. It is closer to the opposite. Better review tools raised the bar after the code was already in, and the project chose to cut features rather than ship them with known defects.
What happened to PostgreSQL 19
PostgreSQL ships one major release a year, usually in the fall. This year the beta cycle produced an unusual number of reverts, meaning features that had been committed were taken back out.
The Register reports that PostgreSQL committer Peter Eisentraut, a vice president at EDB, explained the situation in a blog post dated October 9. He attributed the reverts at least in part to code review assisted by large language models. Some of the pulled features had deep design flaws that would have been hard to fix quickly during beta. Others had a long tail of small edge cases.
According to The Register, Eisentraut said "these kinds of things can get found much faster" with LLM review, and each fix carries its own overhead. The release schedule simply ran out of room. He also said: "The effective quality bar has been raised after the time the feature was written and committed."
He was careful to rule out one popular explanation. The reverts were not caused by "vibe coding," he said, because that style of AI driven coding was not available when most of these features were written.
The PostgreSQL 19 delay by the numbers
A Snowflake engineering blog post from September 16, written by developer advocate Elizabeth Garrett Christensen, put numbers on the trend. It counted 53 reverts since the beta began in June, compared with about 44 for PostgreSQL 18. At that point the post already warned that the release was "certainly delayed by weeks and maybe even months."
The same post noted that the August 2026 PostgreSQL 18 patch release fixed 28 CVEs, far above the usual count. It linked that jump, along with the late surge of bug reports, to AI tools that help find bugs and build reproducible test cases.
Features pulled from PostgreSQL 19
The Snowflake post listed the features reverted as of mid September, including:
- SQL Property Graph Queries (SQL/PGQ), which add graph style queries to the SQL standard
- ALTER TABLE MERGE and SPLIT PARTITION
- UPDATE and DELETE FOR PORTION OF, for temporal ranges
- GROUP BY ALL, which produced wrong results in some cases
- Switching default TOAST compression to lz4
- Nontext output formats for pg_dumpall
- JSON_TABLE ON ERROR cascading
- Online data checksum changes
- Nested query tracking in pg_stat_statements
SQL/PGQ is the headline loss. The feature was added to the SQL standard in 2023, and PostgreSQL's version was removed over unresolved issues. The Register reports that longtime core developer Tom Lane had warned that shipping it risked "post-release bug discoveries that are unfixable until v20."
What still ships
PostgreSQL 19 is far from empty. The Register says REPACK CONCURRENTLY, which reclaims disk space without the heavy table lock that VACUUM FULL needs, remains in the release. The Snowflake post listed other features still expected, including pg_plan_advice, parallel autovacuum, ON CONFLICT DO SELECT, IGNORE NULLS in window functions and several logical replication improvements.
Eisentraut, per The Register, still calls it "a great release," adding: "And it will be very robust, after all the additional scrutiny."
Why AI code review changed the schedule
This is one of the clearest public examples yet of AI review tools changing how a major open source project ships. For years, many of these edge case bugs would have surfaced slowly, through user reports and point releases after launch. Now they arrive in a flood, before release, and every one of them needs a human to understand it, decide on a fix and test it.
Other projects are feeling the same pressure. Google paused part of its open source bug bounty after a wave of AI generated reports, as we covered in our story on that pause. On the other side, AI coding help can deliver real wins when experts steer it, as when Linus Torvalds used Claude Code to rework the Linux page cache. And the risks of moving too fast are real: a recent survey found AI coding tools caused production outages at most large firms.
PostgreSQL's choice sits in the middle. It used AI to look harder, then let people decide what was safe to ship. The Snowflake post also noted that the project has no official policy on AI assisted contributions yet, though the author expects one soon. Robert Haas, another core contributor, has run a bug hunting contest that uses AI assistance.
Who is affected
Most companies running PostgreSQL will not feel much pain. Few teams jump to a new major version on release day anyway.
The groups that will notice:
- Teams waiting on specific features. If you planned around property graph queries, GROUP BY ALL or partition merge and split, those now look like PostgreSQL 20 material.
- Cloud and managed database providers. Services that offer new versions quickly will have to adjust their own launch plans.
- Extension and tool developers. Anyone testing against PostgreSQL 19 betas should check that their code does not rely on reverted features.
What to do now
The practical advice is short. The Snowflake post put it plainly: "PG18 is the safe upgrade target today." If you are on an older version, PostgreSQL 18 is the release to plan for right now.
If you are testing PostgreSQL 19, keep an eye on the pgsql hackers mailing list, where any further changes will be discussed. Update your test plans to drop features that have been reverted, and expect the final release around the November 19 target The Register reported.
For PostgreSQL 20, the picture has changed. Developers had expected it to be a quieter release. It now looks likely to carry many of the features pulled from version 19.
What to watch next
Watch for three things: whether the November target holds, whether any more features get pulled before release and whether the project publishes a formal policy on AI assisted contributions and review. How PostgreSQL handles this may become a template for other large open source projects facing the same flood of AI found bugs.
For more coverage of developer tools and the wider industry, visit our tech section.
Frequently asked questions
When will PostgreSQL 19 be released?
The Register reports a target of around November 19, 2026, after an original plan of late September or early October.
Why was PostgreSQL 19 delayed?
Developers reverted between 8 and 10 significant features. Committer Peter Eisentraut said AI assisted code review found problems faster than they could be fixed before the release.
Which features were removed from PostgreSQL 19?
Reverted features include SQL Property Graph Queries, GROUP BY ALL, partition merge and split, and FOR PORTION OF updates and deletes. Most are expected in PostgreSQL 20.
Should I upgrade to PostgreSQL 18 or wait?
A Snowflake engineering post calls PostgreSQL 18 the safe upgrade target today. Teams that need stability can plan around it now.
Share this story
Spotted an error? Tell us
