I'm seeing something interesting regarding LLMs. I'm seeing some folks use LLMs to push something further than any team of humans would reasonably push it before going insane. They might be pushing perf, memory usage, precision, accuracy... Humans would have gotten 80-90% there, then leave that last 10-20% as "not worth it". LLMs allow people to push things to absurd levels. But pushing things that hard still requires lots of effort, and care. It still requires (gasp!) engineering.