8 Likes
Quite interesting post. I’m not sure if Matt is reading this thread, but lately I’ve been wondering about use of LLMs for code generation.
To me personally, it takes way more time to review a patch to a degree where I could say:
- I understand it fully
- I could replicate it from scratch
…as compared to just writing it myself in the first place. That’s especially true when you just look at a diff, because you’re only seeing a fraction of the context (the control flow).
So I’m wondering if what people do is actually deep code review or more like: the patch does what I want and seems to not break things.
Yes, it’s not the same as vibe coding, but it has implications as well.
2 Likes