When you’re trying to develop a vertically integrated feature across a synchronized release requiring potentially new silicon, a new device, new OS frameworks, new app code… you have to express your requirements precisely. Either the M1 is being designed to support three displays or it’s being designed to support two. Not “as much display support as we can squeeze in where performance is still OK end-to-end”. By the time you know if end-to-end looks good for all the features you built up depending on lower layers in the stack, it’s too late.

You’re also likely not to trust “hey, seems like our tolerances were excessive and it works great on older hardware”. And building up that trust is time-consuming and difficult, so they rarely go back to do it without a strong market justification. Stage Manager being the most recent—somewhat odd—example.

Which is weird because who really cared about stage manager? I guess they decided the media kerfuffle was making owners of unsupported recent hardware feel put down.

So whilst Siri could technically run on older devices the experience was unacceptably poor.

Almost anything can kinda work on older devices. But lots of little details make the difference between a good experience and a poor one. Which simd instructions did it support. What’s the battery impact on that BT chipset. Did the ANE support NN layer style X?

Apple has a great track record of brining new features to old hardware. I don’t see example here or elsewhere that I think were purely greed and not quality driven.

E.g. the apple watches really “inherited” a good chunk of all the new features to the point that there are several versions that are basically identical. Like, I have the 6, and besides the on-screen keyboard (for which I guess the screen was too small based on their testings) and temperature-sensor reliant features, it does almost everything the new 10 will be able to do.