I am just pointing out, ideally this would be done at the root level with machine code, specific to all architectures. A "function model" would return the best algorithm for any natural language prompt (or function prompt requirement).

If anyone is trying to understand it better, I wrote it up here https://circuspam.coffee/2025/05/10/the-future-of-operating-systems-building-a-dynamically-generated-ai-os/