alt Hacker NewsI think more than ever programmers need jobs where performance matters and the naive way the AI does things doesn't cut it. When no one cares about things other than correctness your job turns into AI Slop. The good news right now is that AI tends to produce things that AI struggles to do well with so large scale projects often descend into crap. You can write a C-compiler for $20,000 with an explosive stack of agents, but that C-compiler isn't anywhere close to efficient or performant.
As model costs come down that $20,000 will become a viable number for doing entirely AI-generate coding. So more than ever you don't want to be doing work that the AI is good enough at. Either jobs where performance matters or being able to code the stack of agents needed to produce high quality code in an application context.
Replies
I wonder what other “crevices” (as the author put it) exist.
Another commentor mentioned embedded, and after a brief phase of dabbling in that, mainly with nRF5x micros, I tend to agree. Less training data and obtuse tooling.
> When no one cares about things other than correctness
I don’t get the impression that the majority particularly cares about correctness. In fact, it’s one of the weak points of AI.