PostgreSQL 19 slips to November after AI code review outran the beta window

PostgreSQL 19 will not arrive in late September or early October as planned. It now expects general availability around November 19, after developers reverted between eight and 10 significant features to PostgreSQL 20. Peter Eisentraut, a vice president at EDB and a leading committer, says that count can be attributed at least in part to the LLM-assisted code reviews developers now run.

What the reviewers found

In his October 9 post, The reverts will continue until morale improves, Eisentraut separates two kinds of finding. A few reverted features had significant architectural defects that would have been hard to fix "quickly and during beta". The rest were a long tail of "relatively harmless issues involving various edge cases" that in the past would have surfaced slowly and been patched over the five years after launch. Cheap review turned that tail into a queue: "With LLM-assisted code review, these kinds of things can get found much faster, and then you're staring at a list of forty defect reports." A three-line fix still has to be read, tested and argued about, so per-item overhead does not shrink with the diff.

Why fewer features is the intended outcome

Better review did not shorten the release. It moved the constraint from finding defects to disposing of them. Eisentraut's sharper point is about timing: most of these features were written when LLM review was weaker, so "the effective quality bar has been raised after the time the feature was written and committed." Code that cleared an older bar was re-judged against a newer one, inside a window that did not move.

He also rules out the lazy explanation. The reverts were not "vibe coding", because the technique was not available when most of the features were written. Nor was review the only pressure: senior developers spent April to August on security fixes, thinning the beta review that normally catches such bugs.

What still ships, and what it proves

SQL Property Graph Queries, part of the SQL standard since 2023, was removed over unresolved issues. Contributor Tom Lane had warned in September he would "be willing to bet dinner" that shipping it in v19 would bring bugs unfixable until v20. REPACK CONCURRENTLY, which reclaims disk space without the table-locking VACUUM FULL, stays. Eisentraut calls the result "a great release" that will be "very robust, after all the additional scrutiny". The pulled work heads for PostgreSQL 20, which a few months ago was expected to carry no new features.

The causal claim is his, and he lists bad code and coincidence as candidates too. What the episode does establish is arithmetic. When review finds defects faster than a fixed window can close them, features leave the release. Eisentraut expects the situation to sort itself out unless LLM review keeps improving at its current rate.