PostgreSQL 19 delayed to November 19 as AI finds more bugs

More bugs, no more humans

In a post published on 30 October, core contributor Robert Haas put it bluntly: large language models have arrived. In his view, they find flaws that decades of human review missed, but they do not remove the bottleneck. Every proposed fix must still be understood, validated, adapted to project standards and merged by human maintainers. Bugs are discovered far faster than they can be resolved.

Haas compared code changes made between 8 April and 25 September each year. Changes applied to the upcoming version rose from 724 in 2025 to 1,169 in 2026, a 61% increase. Changes that also touched the previous, already released version climbed from 248 to 433, up 75%. He attributes this less to more contributor mistakes than to more people reporting bugs found with LLMs.

Security fixes are growing too: the 13 August update patched 28 vulnerabilities, against 11 in May. Neither that count nor Haas's figures show how many of these problems were found thanks to AI.

A different problem for open source

Haas considers the current delay modest, but is less calm about the long term. "How are we going to catch up?" he asks, admitting he does not know how the project will keep innovating.

Earlier complaints about AI in open source focused on worthless reports. In January, curl ended its bug bounty to discourage low-quality submissions, and on 1 October Google suspended new submissions to its open source rewards program, citing mostly invalid automated reports. Haas describes something different: real bugs, in numbers developers struggle to handle while preparing a release.

The next milestone is 12 November, with an update expected to carry many fixes and a final test version of PostgreSQL 19. If no problems appear, the final release follows on 19 November.