Post

“LLMs will make codebases into unmaintainable slop!” Ed: “I just had built a ts-morph custom linter for free that enforces business rules the type system cannot. It required zero of my attention and is patterned to add more whenever needed”

8 reposts105 likes

Hot take: instead of making everything slop, LLMs will create a bimodal distribution in quality. On one side, codebases of an extremely high quality, previously impractical to attain. On the other, a level of code quality so low that we'll need to invent a new term for it. My money's on "hyperslop"

I’ve been going in the other direction: working on mistakes in a code base that would require too much jackhammering to do by hand. Not without hating your life anyway.

LLMs are good at spewing (even the case of fumbling around looking for extractions is a form of spewing diff proposals and test commands), but with discipline, the spew can be harnessed.

i think my take on that is claiming your tui tool is more akin to a ‘game engine’ yet you have 11ms frametimes (~90fps) for what you would think should be an easy win

ya this is the part i’ve been thinking a lot about, all the “damn imagine the QOL/UX boost from that tantalizing 3 day rabbit hole” are maybe an hour to critique and review the diff stuff you’d just never take on without infinite time that can be robustly tested is so tractable now

i guess the only question is how many people are going to realize that they can use llms to unlock this sort of productivity boost

This is the kind of thing I've been doing and haven't really talked about it so people don't crucify me

everybody. just takes time. one thing I'm interested in, which I see people pointing out more and more, is that smaller teams can adapt to this much quicker. maybe bigcos are going to have give teams more autonomy if they want to see adoption