Post #1863810

2026-04-30 15:27 UTC

A PostgreSQL rewrite in Rust? Awesome! Reads a little more. Oh, it&#39;s using multiple LLM agents to produce over 450K lines of code that I&#39;m _totally_ sure will be thoroughly reviewed by the *one person* working on the project. I don&#39;t care how many regression tests it passes, it&#39;s going to be a flaming pile of dog poo for the same reasons Anthropic&#39;s C compiler was.

Replies (3)

@peterhuene@hachyderm.io 2026-04-30 15:30

"Yesterday, I added around 90k new lines to pgrust" Do you even hear yourself, my guy? That's a lot of code that the _LLM produced_ that you literally could not have reviewed. There's not enough seconds in a day to be able to review that much slop.

Open ##2831778
@esoterra@hachyderm.io 2026-04-30 15:35

@peterhuene@hachyderm.io yup! like this post argues, each time you fail a test and make adjustments, you use up some of its descriptive power. LLMs do this at *scale* making the tests useless. https://types.pl/@wren6991/116488366700176901

Open ##2831779
@ianthetechie@fosstodon.org 2026-05-01 02:12

@peterhuene@hachyderm.io yeah… I am not actually pessimistic on the idea of LLM assisted rewrites (I’ve done a few myself), but this isn’t the way. You really do have to review every line carefully in critical systems like this. Even with clear requirements they get basic stuff wrong that even a mid level engineer wouldn’t.

Open ##2831781