I think there's probably an underexplored sweet spot where IDEs gain the ability to hint at what type of optimizations the compiler will be able to do, and you

I think there's probably an underexplored sweet spot where IDEs gain the ability to hint at what type of optimizations the compiler will be able to do, and you could even specify which ones you want to select visually. Which in term is left as a textual annotation that's "statically checked", i.e. if the compiler cannot perform the expected optimization anymore it will reject the annotation with an error or warning whichever degree of freedom you decided to give the compiler. Could get a bit unwieldy but it might not have to. You could also combine this with generative ML to suggest refactors that lead to potentially more easily optimizable code. It doesn't have to be perfect just assist enough for you to gradually build intuition. ML could also provide estimated guesses on how an optimization might speed up a given logical unit of code. What you get is writing readable code but still a clear idea of how it will behave / impact the compiler. It does of course go counter the idea of abstraction. But if we do advocate for people to learn how the compiler will optimize or not we might as well invest in tooling to support that.