This writing machine would permit you to use a new process of composing text. For instance, trial drafts could rapidly be composed from re-arranged excerpts of old drafts, together with new words or passages which you stop to type in. Your first draft could represent a free outpouring of thoughts in any order, with the inspection of foregoing thoughts continuously stimulating new considerations and ideas to be entered. If the tangle of thoughts represented by the draft became too complex, you would compile a reordered draft quickly. It would be practical for you to accommodate more complexity in the trails of thought you might build in search of the path that suits your needs.
Modern Day LLMs and such seem to fill a higher abstraction of this process as a tool but the question then is at what parts do you want it to handle the complexity vs human cognition handle it.
As in, is there an abstraction limit to the amount of complexity a human can handle in this way? Up until now it seems there is no limit and we have augmented upwards infinitely with new technology.