[obsolete_wagie]

"AI's out here, a gift from the heavens (or at least from Sam Altman's nerd fortress) ready to write your shitty little to-do app in five seconds flat. It can churn out pixel-perfect HTML, debug your fuck-ups, and probably even wipe your ass if you ask nicely. But no, you're still humping your frameworks like they're the last lifeboat on the Titanic."

Personal opinion is that AI will reduce the need for higher abstraction software libraries. ORMs for instance could go away. We will see wildly different software paradigms as the need for human understanding drops

[ratorx]

I don’t think it’s quite that simple. Taken to the logical extreme, everything above pure machine code is “abstraction”, but I don’t expect AIs to produce good machine code any time soon.

Even if you consider trainability (amount of code etc), Python is a higher abstraction than C and I don’t see that going away either.

A more nuanced view is that libraries that exist to reduce boilerplate will likely see less use, whereas libraries that exist to simplify a problem domain or similar (automatic memory management language, crypto libraries, parallelisation abstractions) will stay, at least whilst we are relying on humans to review AI generated code.

[woodrowbarlow]

i'm not sure i agree. maybe if you're "vibe-coding", but not if you're using AI as an assistant. a good abstraction makes it hard to write bugs, so telling AI to use a certain library (which i know to be high quality) is a good way to constrain the types of bugs i have to look for when reviewing the code.

[obsolete_wagie]

once AI reaches a certain level of accuracy, it will seem foolish to make it use abstractions built for humans

[garylkz]

and once the "accurate" AI accidentally wrote bugs, it will seem even more foolish for humans to debug the binaries built by AI.

[daxfohl]

Or it could make abstractions even more important and useful. If it's able to identify common but nuanced patterns across the entirety of github, it could build libraries and frameworks that satisfy them. Then it would fine tune itself based on those tools, and start using those abstractions going forward.

That would improve the consistency and reduce the complexity of software everywhere in a way that gaggles of human engineers across thousands of computers never could.

It would be beneficial for the AI too, as the fewer things it has to keep track of, the more efficiently and accurately it will be able to generate correct applications on top of it. This would cut costs, hallucinations, and allow smaller local models to perform better.

[dsego]

> Personal opinion is that AI will reduce the need for higher abstraction software libraries.

I'm not sure we should be excited by that. Instead of building more powerful high level abstractions we're giving up and hope to build better software by churning out tons of one-off spaghetti code.

[obsolete_wagie]

this negativity around ai software is misplaced, and will be on the wrong side of history. the trend is ai built software that will be better than human built. vibe coding is just early adoption

[netdevphoenix]

> We will see wildly different software paradigms as the need for human understanding drops

Isn't an LLM basically another abstraction layer? Unlike React or EF Core, it can talk back

[obsolete_wagie]

I think of LLMs as a compiler of language into code, it can generate lower level code than we would normally build

[netdevphoenix]

Makes sense

[mrkeen]

LLMs reinforce what the majority is already doing, because that's what it's trained on. I'd like ORMs to go away but LLMs aren't going to do that.

[obsolete_wagie]

its more fundamental than this, ai can certainly take an orm and replace it with generated sql. it can do the same for react components into html (still early days for this though).