The description of interacting with a computer through language seems oddly prescient of LLMs.

I was thinking the same thing.

I think what it describes is about right too: computers programs should have a REPL and we should have agents that can input them for us if we don't want to get into the weeds and wish to automate tasks, in some ways anticipating the browser too.

This was always the holy grail, we just were never able to achieve this practically.

For example see the early Zork style games, or declarative style languages like SQL. It turns out that it's very tricky to get computers to "understand" language like humans do, and approximating it with pre-defined grammar and vocabulary is not enough.