Post #3823222

@bkuhn@fedi.copyleft.org @next@fedi.copyleft.org Interesting read. Thank you for sharing it here! With apogogies for length, this got me thinking, and I got stuck on one point that I think merits more discussion: Imagining that AI can rewrite any code is doing ourselves - as maintainers - a disservice. AI can remix any code, but getting code *secure*, *correct* and *efficient*, while covering all edge cases is *really* difficult. AI can help with really difficult problems, but the easy hand-wave version of AI has nothing useful to contribute. Across my career - I have seen hundreds of problems that I thought were easy to quickly code perfect solutions for. In every single case, what was really happening was that **I had a deeply naive understanding of the problem space.** AI code duplicates my junior developer naivety and multiplies it's harm through rapid automation. There are certainly plenty of cases where that's still good enough, just like my early code contributions were still good enough for those who hired me. But the narrative that cheap AI remixes will meaningfully and lastingly replace *important code* is just sales hype bullshit that our coding communities seem to keep falling for. Hard problems rarely yield to anything less than hard work. For important code, we have to think about maintenance lifecycles, supply chain protections, and sometimes even a clean title of intellectual property - all of which most forms of AI code assist makes harder to manage, not easier. We know this parable, but it's not new or interesting or exciting enough to get attention, in the current discourse. The rabbit wins trivial sprints. The tortoise wins marathons while the rabbit is busy celebrating prematurely.