Interesting. The reason why companies aren't trying their best yet into non-static weights/online learning is probably (cloud) logistics. It seems simpler, easier and cheaper to serve a static, well-evaluated, and tuned model, rather than trying to let it learn alongside a specific user or all users.

Oh absolutely. To be clear... I think this is probably a bad idea. It probably wouldn't be successful and if it was you'd have very little control of how it evolves.