With the expansion of AI-based artworks in public spaces, the conceptual and structural analysis of this form of art within urban contexts has become a pressing research concern. Accordingly, this study aims to propose a framework for classifying AI artworks in public art. The main research question is: according to what criteria can AI-based artworks in public spaces be categorized, and how can such a classification contribute to understanding the audience’s experience and the role of the intelligent agent? This qualitative study adopts a descriptive-analytical method and is based on the analysis of nine significant examples of AI artworks presented in urban environments. Drawing on Ahmed’s theoretical framework (2018) on interaction and interactivity, and through a reinterpretation and development of Mendelowitz’s conceptual model (2020), the study formulates a five-part model consisting of generative, reactive, interactive, learning, and static works. These categories are organized according to the type and operational behavior of intelligent agents, the mode of interaction, and the audience’s perceptual experience. The analysis of the selected examples shows that the level of agency, the degree of responsiveness, the source of input data, the stability or variability of system behavior, and the form of audience experience vary significantly across these works. Static and generative works offer a more passive experience that remains independent of the audience, whereas reactive works respond to environmental stimuli, and interactive and learning works, through the use of adaptive and learning algorithms, enable dynamic, participatory, and customizable experiences. The findings indicate that the final classification of each work depends not only on the type of intelligent agent, but also on the relation among system, audience, environment, data, and interaction. The conceptual model proposed in this article can be used both as an analytical tool for understanding the structure of existing works and as a framework for designing future artworks within the context of smart cities.