Boy, are you going to be blown away by Pac-Man.
Similarly, when the emulator broadcasts current coordinates for pieces, or sprites of them - they still can be reduced to plain text.
As a matter of fact, if you want to play Pacman in Emacs... you can just play Pacman in Emacs: https://github.com/emacsmirror/pacmacs
It's like arguing that "all books are just sequences of letters". That's technically true, but meaningless without understanding grammar, semantics, and context.
What I'm saying is akin to: "First step to understanding all the books starts with understanding letters..."
Anyway, emojis are of course text - even though they are stored as encodings. For example, LLMs process them exactly like any other text characters - they're tokenized, embedded, and handled through the same mechanisms as words. I should've use a different analogy to make my point clear, but you know what I meant.
What do you mean by "non-text things"? You are dealing with the computer, it's all about text and mostly text. Sure, it might be encoded and digitized, but even structured binary - is all just text. Even when you give a computer voice commands - they are just synthesized audio form of fucking text. Even with "turtles all the way down" - it's all turtles made of text.
Have you seen the gif in the blogpost? With the transient that has "Move Forward/Backward", "Increase", etc. commands? The commands that you'd have to send to the specialized app anyway - Emacs or not. In what form? Fucking text, of course.