Of course, it's not like type stalking doesn't have benefits for the customers. Many people won't mind finishing up a customer service chat as efficiently as possible. But is it actually efficient? Socca also noted that the quick-draw answer he received was barely relevant to the question he wound up asking, likely because the agent, like a shitty human autocomplete, had written their reply based on their guesswork of what Socca would wind up "sending."
And if the system is so great and uncontroversial, why hide it? Why even still have a send button at all? Why give the customer the illusion that they decide what gets shared? If they have to disclose to customers that their calls are being recorded for training purposes, they could easily start off every chat with the disclaimer "Our agents can see you typing 'balls' in a long uninterrupted string after misspelling 'definitely' on your fourth attempt."
So how do you regain your privacy when talking to a customer rep? Easy: Type messages in a note-taking app or a word processor, so that only you and whatever third parties the software sells your identity to will read it (which is the most intimate the internet will ever get). Or you can just put on a tinfoil hat, scribble it down on a piece of paper, and then type it with the furious alacrity of a conspiracy theorist who knows everyone is watching. Sure, it'll take longer, but you were already queuing 45 minutes to complain about your toaster's refund policy anyway, so what's another five if it will the soothe every bloating paranoia?