> what happens when AOSP gets a computationally expensive update (i.e. embedded ML voice recognition or realistic TTS)? You disable those features on lower-spe

> what happens when AOSP gets a computationally expensive update (i.e. embedded ML voice recognition or realistic TTS)? You disable those features on lower-spec phones or fall back to a low-resource alternative. If I'm on an older device that can only run eSpeak instead of a more realistic but expensive TTS engine? Fine, I'll run eSpeak! You don't have to bundle everything together into an image for each device. If you want to disable stuff, do it at runtime. Or if you really need to remove stuff due to storage limitations, have a small manifest that can be used to build an image with unused things excluded. This doesn't have to be "oh we need to make completely custom versions of the OS for each device", this is basically "install these packages, don't install these ones". Modularity!