Your first sentence is basically impossible. When reviewing code, your insight is always inferior to the person who wrote it. Often my code reviews consist of questioning the programmer on why they’ve done what they have, because I need their perspective as well.
You can’t do that with an LLM because it doesn’t really know why it did things.
And testing properly…? Nobody does that. LLM or human.
I was warming up to this. It certainly is better than vibe coding, but I’ve started to see libraries [with this ai policy] push out bigger releases faster with bad ergonomics and unnecessary, undocumented breaking changes, because the ai quietly migrated the internal tests/examples.
Clearly, they’re reviewing large amounts with an eye on new features and they’re not thinking through ergonomics: spending time with it, writing it themselves, and seeing where the sticking points are.
Maybe if you’re using ai for the migration, then-- problem avoided, for now. As time goes on I could see it compounding and becoming a tangled mess that’s not only inaccessible to humans, but requires more and more compute to navigate than thoughtful, handcrafted code.
Your first sentence is basically impossible. When reviewing code, your insight is always inferior to the person who wrote it. Often my code reviews consist of questioning the programmer on why they’ve done what they have, because I need their perspective as well.
You can’t do that with an LLM because it doesn’t really know why it did things.
And testing properly…? Nobody does that. LLM or human.
I disagree with this. Just reviewing the code is exactly what I mean by “properly being checked and tested”.
There are people/companies that just vibe code the hell out of their stuff without any review whatsoever. That’s not something I want to touch.
I was warming up to this. It certainly is better than vibe coding, but I’ve started to see libraries [with this ai policy] push out bigger releases faster with bad ergonomics and unnecessary, undocumented breaking changes, because the ai quietly migrated the internal tests/examples.
Clearly, they’re reviewing large amounts with an eye on new features and they’re not thinking through ergonomics: spending time with it, writing it themselves, and seeing where the sticking points are.
Maybe if you’re using ai for the migration, then-- problem avoided, for now. As time goes on I could see it compounding and becoming a tangled mess that’s not only inaccessible to humans, but requires more and more compute to navigate than thoughtful, handcrafted code.