AI demands more engineering discipline. Not less

(charity.wtf)

7 points | by RyeCombinator 2 hours ago ago

1 comments

  • mikgp an hour ago ago

    > Code becomes precious when it is the only place knowledge lives.

    It seems like a lot of people amidst this AI revolution want to flip "Working code over comprehensive documentation" on it's head. And, maybe I'm not that smart, but this feels wrong to me.

    It's a weird middle ground to find, because so much code is basically boiler plate. But this idea that "no, it's ok, knowledge can live anywhere in any format" that we're going to engineer off of a mediocrely maintained wiki page over code which is an authoritative source of truth is bonkers.

    I don't know if the answer is spec driven development exactly, but it does feel like some kind of higher written language with broadly accepted semantics is required to continue calling engineering a discipline?

    > I am not asserting that all code will eventually be AI-generated to spec, bypassing human understanding.

    Why not? Because on the other side, it's a weird middle ground to continue to have to review code. To have to piece together truth based on a few lines of programming language you read alongside with a spec that is defined? Whatever thought leadership needs to happen to get us to this new world, it doesn't seem entirely complete.

    This article is heading in the right directions. She kinda gets there with the immutable infra perspective, but immutable infra largely works as a collection of agreed upon specs - terraform code bash scripts, etc.

    And this is where it gets weird because in a sense a library, or a framework is just a spec; and none of us can agree on a standard one of those.