Why would this be the case when the text that produces binaries (code) is usually both more token efficient and vastly more effectively organized for modification/extension?

Unless by bits you just mean text in general, or any data since it’s all bits, in which case what you’re saying is trivially already true.

It seems like you’re saying that long term LLMs will output pure machine code as the most effective way to use them.

Or just replace all of the apps? Why would a user use apps if everything (eventually) can be achieved with an AI?

well they can natively converse in base64

are you implying that text is impure bits?