As usual with Alan Kay, it's difficult to understand exactly what he's advocating for. On the one hand, he laments browsers having features, and thinks they should just execute code safely and provide a low level graphics interface, but on the other hand, he advocates for a universal polymorphic graphical object mesh (perhaps approximated by LivelyWeb) which would require a great deal more "features" (think Java) and is closer to what OP is describing.

I think he's made his way into this thread, so I invite him to clarify!

I read it as the browser should just be a VM. It's low level. Then the objects that run in the VM provide the features. This solves a lot of problems. The browser is slowly evolving to do this anyway with js, then wasm (reality is proving Kay right) but the underlying high levelness of it makes it a mess. A VM is more flexible and robust and at the same time simpler than a browser

The difference is that, in order to foster composition, the platform can't just be a VM. It must also provide the interfaces by which objects interact. Take Java. It's not just a VM. It's also the JDK/JCL.