> > You can't trust the AI's code! > Yeah, but you can't trust your own code! Those are two different types of trust. The only overlap they have is that they

> > You can't trust the AI's code! > Yeah, but you can't trust your own code! Those are two different types of trust. The only overlap they have is that they probably both have bugs. Maybe I'd trust AI code after some really aggressively adversarial TDD and fuzz testing enough to have a tiny bit of hope that it's going to do the right thing in most cases. But for anything non-trivial I'd be worried about a bunch of other trust issues. Just off the top of my head: - You know all those articles with terrible advice Google is always showing at the top of their results? Welcome to a world where those are regurgitated by a well-articulated, authoritative program, resulting in a long-term maintenance nightmare. - An LLM is far too big to review for malicious content. And if it's being used by lots of programmers you can bet a billion it'll be a prime target for bad actors. - The interaction between my brain and my hands isn't going to be shipped off to professional manipulators.