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

PostgreSQL 19 is expected to arrive in November 2026, later than the project’s usual mid-September-to-mid-October window. Several substantial features - including SQL property graph queries, portion-based updates and deletes, and partition merge/split commands - were also removed late in the beta period. Robert Haas argues that the unusual delay and reverts reflect a surge in bugs uncovered with large language models, rather than a sudden decline in developer care or code quality.

Commit counts support the case for heavier maintenance: between feature freeze and September 25, commits to the release branch rose from 724 for PostgreSQL 18 in 2025 to 1,169 for PostgreSQL 19 in 2026, while fixes to the previous release branch increased from 248 to 433. LLMs can help identify defects and propose patches, but their fixes still require human review; meanwhile, maintainers’ attention has shifted toward repairing bugs and security issues in released versions, leaving less time to assess new features. Haas considers the release delay a modest, quality-driven cost, but sees a harder long-term problem: AI may expose decades of defects faster than maintainers can fix them, potentially slowing innovation. He views this scrutiny as both a challenge and an opportunity that will require software developers to raise their standards.