Everyone has some sort of category error to an extent. But more than that, the part about safety guardrails really resonates with my experience. I inevitably have to go down to the low-level to inspect memory, but because of safety issues, I can barely use Codex as it keeps prompting me for 'trusted access.' AI is actually really good at catching bugs, so I'd love to use it, but it does this every single time...

> I can barely use Codex as it keeps prompting me for 'trusted access.

If you run it in unsafe mode, it won't prompt. The problem is that you would probably want to run it in a container and that's annoying.