logoalt Hacker News

dathinab • yesterday at 8:08 PM • 0 replies • view on HN

Honestly there are 2 overlapping problems:

1. go fast and break things became the norm not the rare exception, mostly due to under-regulation when it comes to responsibility and other much more complex side effects of what you can describe as the US forgoing a fair free marked in favor of out of control monopoly markets they can use as soft power against other countries

2. software being really hard and most tries to fix that being stuck dead before maturing due to them being unfamiliar and often complex to get started with. There being a huge problem with opposite bad ends of "misguided simplicity" and "programmers loving 'clever' & overengineered" didn't help exactly either.

> It wasn’t always like this.

no it was, yes the 2010 felt like a high time. A lot of 90th nonsense has finally been left behind (but would still lead to security issues years later, e.g. log4shell in 2021); Tools for analysis had gotten better; Compute allowed expansive test suites and "go fast and break things" hadn't yet been spread to widely to critical infrastructure (at least kinda, _a lot of the issues with Azure today root back to decisions done in the 2010th_!!).

Most importantly while having well educated highly skilled Devs is needed, _it isn't sufficient_. They still are human. Blaming the huge flood of CVEs on AI generated code is simple, but wrong. Many of the CVE in recent years are human caused, some older then usable AI code gen and at least in case of the Linux kernel much of the code of the past had been written and reviewed by pretty professional, skilled, well educated developers and still AI found way way to many CVEs in it.

So yeah IMHO there is a long standing fundamental issue with how to handle software production, starting with the very first step of requirements engineering (which often isn't done, or at least not properly) and ending with how we fundamentally compose software components.

And the problem I see with AI it makes it worse. We already where approaching a breaking point with software dev and I hoped that this would create pressure to finally find a working way to reinvent it from scratch. But now AI allows us to apply the same failing approaches even more large scale, faster, cheaper and sure with more bugs, but also with AI making many bug fixes far faster and cheaper (if correctly used, it often isn't).