Instead, Google is *starting* from the premise that it wants you to use an LLM, then the feature (generating a message based on summary and tone). The actual use case is added after-the-fact, begrudgingly, because real human beings looked at the feature and shrugged.
That's why it's shit.