alt Hacker News> Adversarial review loops of N parallel agents looking at whatever characteristics you care about will make it better
Especially if different model providers. It becomes more like having team members that see things slightly differently.
> Adversarial review loops of N parallel agents looking at whatever characteristics you care about will make it better
Especially if different model providers. It becomes more like having team members that see things slightly differently.
It's like a variant of the halting problem though: given N agents in a review loop reviewing a codebase, will it ever terminate and say the thing is done and bug free?
It seems to me just from @codex review, given a codebase of any appreciable size, if you ask the agent to find things to fix, it'll find things to fix. And despite N agents agreeing that some code is ready to ship, I've still sat down to try it and nothing actually works as advertised.
So the need for better verification tools that are not AI is very urgent. Because the AI can be told to write a thing without mistakes, it does write code that compiles, N-agent AI review process fixes bugs and eventually the code is approved, and then run against an extensive suite of tests to prove certain functionality which all pass... and yet it still can be the case that nothing actually works in production.