Almost every dev is focusing their library interfaces to be used by AIs, not humans. That's gotta be some major shift in computer science. π€
Discussion
Thatβs not going to work out well for them.
why not?
AI is incapable of understanding system architecture because it is based in digital computation, unlike human consciousness. Like quantum computers, human brains can process infinite possibilities at once with the context of their ~2.5 petabyte memories. AI cannot ever approach human intelligence until we invent quantum AI or a better analog computer.
So to design things for AI is to hinder the human developers that will come and fix the AI slop in the future, which will no doubt break constantly until vibe coding is rejected as a whole. Things will have to be rolled back to before AI enshittification
LLMs are fundamentally a simplified version of what our brains do. AI will improve by increasing its vocabulary, but the fact is that since human cognition is analog, our brains have infinite vocabularies. The way an LLM works is fundamentally quantizing this process so that it can be digitalized. It turns words (analog) into understanding (for it, digital, for us, still analog subconsciously, if digital in our waking conscious)
Sorry if this is incomprehensible
If the AIs in question are centralized, that seems to me like a new form of dependency...
the race is on, and few realise
It's a crap
i'm not. but documentation, yes, the AI needs that or you watch it scratching around for models to build on.
AIs are dumber than humans so on net i think this is a good change. more docs, more tests, more results faster.
helps humans too. we need user stories and definitions of done, no less than the AI