The most convincing systems do not simply answer. They sense the shape of a moment and answer accordingly.
A response can be correct and still feel wrong
Most language systems are judged by whether they produce the right words. That is important, but it is only half the interaction. A quick, flat answer to a delicate question can be technically perfect and emotionally absent.
People read meaning from pace, emphasis, silence, and the small changes in a voice that say: I understand this matters. When those signals disappear, even a useful answer can feel like it came from a machine standing at a distance.
Emotion is not an effect layer
It is tempting to treat warmth or excitement as a visual setting added after the fact. In a real conversation, however, expression is part of the reasoning. It changes when a person needs reassurance, when they are in a hurry, and when the next useful move is to make space instead of filling it.
That is why we model expression alongside intent and context. The system decides not only what to say, but what it should feel like to hear it right now.
Building for the moments between words
Our work begins with the small human signals that are easy to overlook: a pause before a difficult answer, a softer pace when introducing something unfamiliar, a rise in energy when an idea lands. None of these are theatrical. They are cues that help people stay oriented.
The goal is not a system that performs a personality. It is a system that can be present enough for people to focus on the conversation itself.
Keep a good signal close.
New research and product notes from Sunesis, sent occasionally and thoughtfully.
Subscribe to the journal