The article argues for a genre of AI capable of building social bonds with humans. The argument’s starting point is the two competing origin stories of Artificial Intelligence. In one, the goal of AI was to create machines that could simulate every aspect of human intelligence. In the other, it was to build machines that adapt closely to natural human behaviour. While the first story is better known, it is argued that the second would have been more fruitful, as it places the human at the heart of the endeavour. Based on this historical perspective, the article provides several examples of conversational agents that engage in this kind of adaptive social behaviour. Results of experiments with these social agents find that they do in fact improve relations between people and the systems. Additionally, they improve performance on the task that the human and the conversational agent are conducting together.