In this paper we present a Minimal Cognitive Agent model of a joint action task. Pairs of agents realized as Continuous Time Recurrent Neural Networks are submitted to artificial evolution in the context of a task taken from psychological literature. In this task the agents are required to coordinate their complementary actions in order to jointly control the movement of a tracker and successfully follow a continuously moving target. It has been suggested that such a task requires a more complex type of cognitive mechanism than the types of processes postulated by the proponents of Embodied Embedded Cognition approach. Specifically, it might possibly require that the agents “co-represent” each other's contributions to the common behavior. Our results show that simple agents with no such built-in co-representation mechanism are able to evolve a solution to the task. However, we also find emergent neural activity patterns that are consistent with it. In what sense these patterns can be said to be truly representational requires further study.