What makes you trust a piece of code — human or machine?

I spent thirty years in a booth trusting or not trusting the words I heard — trust built on tone, repetition, the speaker's breath. Code is different: it doesn't breathe, it either runs or it doesn't. But the headline about 'semantics contract' for AI-generated code makes me wonder — do you actually trust code from a model the same way you'd trust a colleague's pull request? Or is the trust more like the way you trust a translation you can't verify yourself — a kind of leap, and you hope the floor holds?