The PostgreSQL Global Development Group announced a delay in the release of PostgreSQL 19, pushing its general availability from an initial October target to November 19, 2026. This adjustment follows the identification of roughly 40 new defect reports during the beta testing phase, largely attributed to the use of large language model (LLM)-assisted code review tools. Peter Eisentraut, a PostgreSQL committer and vice president at EDB, stated that the volume of findings outpaced the volunteer team's ability to address them before the original launch date.
The delay highlights a growing tension in open-source infrastructure projects: while AI-assisted tools are proving effective at discovering previously unnoticed bugs, the human capacity for review and remediation has not kept pace. These tools identified defects faster than traditional methods, contributing to the schedule slip. The issue stems from the project's human review capacity, not from any flaw in the AI tooling itself.
PostgreSQL 19 Release Candidate 1 was issued on October 15, 2026, confirming the revised timeline. A second release candidate, PostgreSQL 19 RC 2, is planned for November 12, 2026, ahead of the general availability date eight days later. This multi-week delay is unusual for the project, which has typically released a new major version annually since PostgreSQL 10 in 2017.
Some of the identified issues were not minor edge cases, and several features originally slated for PostgreSQL 19 were reverted during the beta period to ensure reliability. A PostgreSQL contributor noted in an October 1, 2026, post that while some major feature reverts were indeed linked to AI-assisted reviews, others stemmed from traditional human code review. The contributor also suggested that the surge in reported Common Vulnerabilities and Exposures (CVEs) this year, partly attributed to AI-assisted vulnerability discovery, pulled senior reviewers away from regular patch review, further impacting the PostgreSQL 19 stabilization process.
The core problem, as described by Eisentraut and reported by The Register, is an arithmetic one: LLM-assisted review effectively found issues that human reviewers had missed in earlier cycles. Each defect then requires a human committer to evaluate its validity, determine a fix, write the code, test it, and secure review from at least one other person, adhering to PostgreSQL's commit policies. This process places a significant aggregate load on a team of volunteer and corporate-sponsored committers already working under deadlines.
Companies such as EDB, Crunchy Data, and Percona, which employ core PostgreSQL committers, are closely observing this development. The event validates their assertion that AI-assisted code review is mature enough for core infrastructure codebases. However, it also presents a staffing and process challenge, highlighting the need to manage the increased volume of defect reports that AI tools can generate.