Yeah - in my own personal experience building AI products this still holds true. And you can see it in the way they behave themselves, they don't really write prompts for themselves to use in the future (say CLAUDE.md) when they need the output to be deterministic. They write code for it, then write tests around that code, then fix the bugs in that code.

Yes it all started with "natural language", but even AI has an understanding that you need more structure than natural language when you want deterministic results.