Postgres 19 is late, and it's AI's fault

PostgreSQL 19, typically shipped between mid-September and mid-October, is now expected to arrive in November or later, with three major features—UPDATE/DELETE FOR PORTION OF, SQL Property Graph Queries (SQL/PGQ), and ALTER TABLE … MERGE/SPLIT PARTITION—reverted late in the beta period between August 27 and September 15. The delay stems from a surge in bug reports generated by people using large language models to probe the codebase, which has driven a more than 60% jump in back-patched commits to the release in progress (1,169 to REL_19_STABLE versus 724 to REL_18_STABLE in the same April 8–September 25 window a year earlier) and a 74% increase in back-patches to the prior release (433 commits to REL_18_STABLE versus 248 to REL_17_STABLE). While commits to the master branch between feature freeze (April 8) and September 25 actually declined slightly year over year (1,443 versus 1,037), the volume of LLM-surfaced defects has kept committers busy fixing rather than shipping. The episode illustrates how AI tooling is compressing the bug-discovery cycle for even mature, decades-old open-source infrastructure that underpins a large share of enterprise and cloud database workloads. Watch for whether the PostgreSQL community further postpones the release, whether the reverted features return in PostgreSQL 20, and whether LLM-driven bug reporting continues to accelerate patching across the project's stable branches.