jlawrence6809

In my view meetings are the most powerful channel for social signaling which unfortunately is at least half the revealed job in software engineering. That was (and still is) the hardest thing for me to accept as the child of blue collar parents.

Personally if I had to look back at the rate of bugs of a similar severity my own company produced I think it was higher pre-llm coding and testing.

Yeah no one can predict the future but betting on progress tends to be a better heuristic with technology, especially in AI given the rate of current progress. And as far as I know the availability of (human generated) training data isn't a strict requirement for further progress given test time scaling. The way I see it is we're still running these models on hardware mostly designed before the LLM revolution so we'll be seeing massive gains in, at the very least, the speed and cost of the current frontier models in the coming years which will make all sorts of things possible. I also think it's more likely than not they will keep improving in capability at some rate and the things software engineers are complaining about now will be seen by the labs and worked on deliberately if simply scaling the models itself isn't enough.

Why not?

How is this qualitatively worse than the other single sources of failure we already live with (ie cloudflare)?

Wait until you see the crappy codebases people were generating before AI.

If AI progress stops right now or stalls out soon sure, but if we really do have something significantly better coming, up to and including AGI, I think none of this will be an issue.

Really? I've found LLMs to be really thorough UI testers.

And when something the LLM wrote goes off the rails sometimes you have to look at code to debug too. But you don't need to make reviewing the code standard practice anymore than reviewing assembly code before shipping.

What makes you think people aren't "reviewing" what they ship to customers just because they don't read the code? If the testing goes well that may be all the "review" required.

99% of software out there is a huge ball of mud even if the senior/architect/<insert egotistical title here> started out by building it with the sexiest abstractions. Things just tend to rot and complexify. Software engineers never figured out a way to prevent that, though it is easy to be fooled by the preening and pontificating we do to make ourselves seem like the "very serious professional engineer who cares about code quality very much". I'd keep using AI if it's working for you.

It seems unrealistic protect our jobs by keeping it a secret if LLMs really can do them (still a big if). Maybe if we as an industry had formed unions but it seems a bit late for that.

Okay my tinker-lust has been aroused! Can someone tell me what I can make with this thing so I have an excuse to build it?

If Linus Torvalds is being outperformed what hope is there for the rest of us?

I had no idea KVMs existed until an agent filled me in last night (I had the idea of building one with an fpga so an agent could drive os installation). Lo and behold the next day KVMs are on the front page of Hacker News!