This interdisciplinary study explores the role of unexpectedness in improvisational theatre through the lens of artificial intelligence. Drawing on the concept of the “circle of expectations” (Keith Johnstone), we designed an artificial improv agent capable of generating dialogue that either conforms to, slightly deviates from, or radically breaks conventional expectations at each single turn. Our central question: What pattern of expectedness best supports an engaging and creative improvisational exchange? We hypothesized that consistently predictable responses would lead to boredom and disengagement, while frequent disruption would cause stress and break the flow of dialogue. Instead, we posited the existence of a “sweet spot” where occasional surprises enhance creativity without overwhelming the performer. To investigate this, we conducted a series of experiments in which professional improvisers interacted with AI-driven bots programmed with eight distinct patterns of expectedness. These patterns varied the sequencing of expected, surprising, and disruptive lines to examine their impact on the quality and flow of interaction. Preliminary results suggest that certain patterns of unexpectedness do indeed foster more dynamic and engaging exchanges. Additionally, we observed that individual improvisers employed consistent personal strategies for handling unpredictability, pointing toward promising directions for future research into adaptive co-creative AI systems. Beyond theatrical applications, such a system could also serve as a tool for assessing an individual’s capacity to handle unexpectedness—an essential skill not only in improvisation, but in any context that demands creativity and adaptability.