During a eleven-week installation of an immersive augmented reality (AR) experience at an art and technology gallery, we faced the practical challenges of preparing for live public audiences and of training and motivating a group of nine modestly-paid undergraduate museum docents to not only usher the show, but to perform as “wizards”. The docents played an integral part of the AR Façade experience, replacing imperfect recognition technology to achieve speech and gesture interaction for the players. Our investigation of the wizard docents revealed insights into two conceptually different wizard-of-oz implementations. The docents initially resisted the interface option that required more cognitive load, but after learning how to make it work, many preferred that interface because it provided them more agency within the system and freedom to diversify the experience for audiences.