Hollywood movies such as Bladerunner 2049 and Ex Machina present visions of virtual agents that can charm us, flirt with us, reason with us and deceive us.  But what would it take to actually build machines with these capacities?  I will begin with a rather pessimistic viewpoint, suggesting that we are a long way from building human-like machines, and that we actually know very little about how humans interact.  Without a robust understanding of human real-world social interaction, it will be very challenging to build virtual agents that can interact.  To move forward, I suggest a recipe for building artificial agents and provide an example from our lab in the domain of mimicry.  Our data shows how the detailed study of human mimicry behaviour can translate into an algorithm for building realistic mimicry behaviour into virtual agents.  I will consider what these kinds of results could mean for our theories of the computational processes underlying interactions in both humans and virtual agents.  I will leave the question of whether we should try to build the agents that the movies show us as one for the audience to consider.