I am talking about people starting new projects, which was the subject of this subthread. I am aware that Microsoft is using LLMs to work on Windows, and now they have remote code execution in MS Paint. Great. The ability to commit shit code to something that is already in production is not exactly revolutionary either.

> how PostgreSQL is having dozens of bugs discovered and fixed via LLMs

The PostgreSQL article was specifically about how the LLMs were incapable of fixing bugs, meaning they had to delay their next major version because humans were unable to deal with the volume of issues reported. I did not make a claim that LLMs are useless, could not help review code, etc, but there is a difference between a useful tool and something that generates production-ready software wholesale.

This is what "100" looks like if you try to generate a DB from scratch: https://jepsen.io/analyses/particledb-1.1.4 - written in Rust, ""formally verified"", scammed AI-pilled investors out of $90 million for a database that fundamentally doesn't work in any capacity whatsoever.

>I am talking about people starting new projects, which was the subject of this subthread.

The subthread was also about 80-100. This seems like a goal post move.

>The PostgreSQL article was specifically about how the LLMs were incapable of fixing bugs

We must have read different articles. This one does not say that, as best as I can tell: https://www.theregister.com/databases/2026/10/16/postgres-19...