AI finds bugs faster than humans can verify them

Postgres 19 is late because so many more bugs are being found now. The top comment names the real issue: 'scale verification, because discovery isn't the bottleneck anymore.' Another commenter says AI is great at read-only search (find vulnerabilities) but weaker at write tasks like adding features, because search results are easy to check. The AFL++ fuzzing tips thread adds a related story: one engineer found multiple issues with AFL in code written by people with the highest engineering titles at a FAANG.

The pattern: automated discovery is getting cheap for both defenders and attackers, while fixing, triaging and reviewing stays human and slow. Trigger.dev's exposure of customer environment variables shows the other side. Commenters were angry that the status page said 'Resolved' after secrets leaked, and asked whether the companies hit in these incidents will ever disclose how access was gained.

Nuance: one commenter points out they found major bugs in late Postgres betas long before AI. The volume changed, not the kind of problem.