← Back to feed

PostgreSQL 19 will not ship until around November 19, pushed from a late September or early October window after an unusual number of features were reverted. Between 8 and 10 significant features have been dropped back to next year’s PostgreSQL 20 release.

The cause is not vibe coding. The official release team attributed the GA slip to incorporating a high volume of back-branch fixes, which required a second release candidate. But the feature reverts — the 8 to 10 significant pieces of work pulled back to PG20 — have a separate explanation. Peter Eisentraut, a leading PostgreSQL committer and VP at EDB, wrote in a blog post dated October 9 that some features had genuine architectural defects serious enough to warrant pulling, and that LLM-assisted code review had uncovered a separate long tail of edge-case bugs that would, in earlier release cycles, have gone unnoticed until post-launch patches years later.

The bottleneck is throughput. Eisentraut framed it directly: with LLM-assisted review, a developer can find forty defect reports in a session. Even if each fix averages three lines of code, the queue still requires forty separate human decisions, context switches, reviews, and commits. Volunteer maintainer capacity does not scale with AI review velocity.

The New Open-Source Bottleneck

The PostgreSQL case is the first high-profile instance of an open-source project explicitly attributing a release delay to AI tooling improving bug discovery faster than human maintainers can absorb the output. It will not be the last.

The dynamics are straightforward: AI code review operates at a different order of magnitude than human review. A model scanning a large patch can surface dozens of edge cases in the time it takes a committer to triage one. If the fix rate stays constant and the discovery rate accelerates, the backlog grows.

For commercial codebases this is manageable — engineering capacity can be scaled in proportion to the bug queue. For open-source projects it is more constrained. PostgreSQL has been in continuous development since 1996 and runs a large fraction of the world’s relational databases. Most core committers are salaried engineers at database vendors and cloud providers — EDB, Microsoft, Crunchy Data, Amazon — rather than independent volunteers, but the committer pool is still finite and does not scale linearly with review throughput.

What Dropped

The 8-10 reverted features are scheduled for PostgreSQL 20, the 2027 release. Eisentraut identified the affected work as having architectural defects that could not be fixed quickly during a beta cycle, separate from the edge-case tail uncovered by LLM review.

PostgreSQL’s release cadence has historically produced one major version per year. A six-to-eight week slip is not catastrophic, but it is unusual for a project that last missed its annual window in a comparable way over a decade ago.

The Broader Pattern

The PostgreSQL delay follows a wave of incidents where AI tools altered the risk surface of large software systems rather than eliminating it. Anthropic’s Mythos model prompted Bank of America to bring in roughly 1,000 contractors to patch vulnerabilities it surfaced in internal systems. The CVE disclosure rate has tracked the acceleration of AI-assisted security review.

In each case the pattern is the same: AI scales discovery, humans remain the constraint on resolution. Open-source infrastructure that runs on volunteer labor is structurally most exposed to the gap.

PostgreSQL 19 expected GA: November 19, 2026.