AI Testing Finds What Decades of Human Review Missed — and No One Knows What to Do About It
How this was made Verified AI
Every Intellegix briefing is generated from that day's broadcast and run through automated checks before it publishes — with a human paged on any flag. Here is the trail for this edition.
PostgreSQL 19 is running significantly behind schedule, and the reason is unsettling for anyone who relies on the assumption that mature software is stable software. Robert Haas, a senior Postgres contributor, published a blog post explaining that the adoption of AI-assisted testing tools has dramatically increased the project's bug-discovery rate — faster than the release process can absorb.
The implication cuts deeper than a delayed release date. PostgreSQL has been in continuous development since the early 1990s and powers critical infrastructure at tens of thousands of companies. If AI fuzzing is finding bugs that three decades of expert human review missed, the relevant question is not just whether Postgres 19 will be good when it ships — it's what analogous bugs exist in versions 17 and 18, already deployed in production.
The HN community split predictably. One camp read the news as alarm: if these tools expose systematic blind spots in expert human review of Postgres, every other piece of critical software carries the same unknown risk. The other camp argued the system is working correctly — finding bugs before they ship is the point of testing, and the delay is a sign of quality control, not failure. Both positions have merit, though the alarm case is harder to dismiss.
Separately, a mathematician referred to in the HN discussion as Terry Tao published an essay titled 'Mathematics Has Changed, Let’s Act Like It,' which drew 370 comments — extraordinary for a piece on mathematical epistemology. The essay argued that AI-assisted proof verification tools like Lean and Coq are creating a new standard of rigor that the field's institutions — journals, peer review, training programs — have not adapted to. The HN thread pushed back on what several commenters called a creeping instrumentalism: a proof that no human fully understands, one commenter argued, is not a proof in any sense that matters to mathematical progress. The essay frames formal verification as a scaffold for human understanding, not a replacement — but the thread suggested the community is not fully convinced the distinction will hold in practice.