Computational thinking (CT) is gaining recognition as an important skill set for students, both in computer science and other disciplines. Digital computer games have proven to be attractive and engaging for fostering CT. Even though there are a number of promising studies of games that teach CT, most of these do not consider whether students are learning CT skills or adapt to individual players’ needs. Instead, they boost theoretical knowledge and promote student motivation in CT by usually following a computer-assisted instruction concept that is predefined and rigid, offering no adaptability to each student. To overcome such problems, by benefiting from a probabilistic model that deals with uncertainty, Bayesian Network (BN), we propose an adaptive CT game called AutoThinking. It seeks to engage players through personalized and fun game play while offering timely visualized hints, feedback, and tutorials which cues players to learn skills and concepts tailored to their abilities. The application of BN to AutoThinking not only adaptively provides multiple descriptions of learning materials (by offering adaptive textual, graphical, and video tutorials), similar to the natural way that teachers use in classrooms, but also creatively integrates adaptivity within gameplay by directing the cats (non-player characters) to a specific zone on the game according to players’ ability. Consequently, these adaptive features enable AutoThinking to engage players in an individually tailored gameplay and instill CT concepts and skills.