This doctoral research examines how AI chatbots function in mathematics education through a main research strand about students’ interactions with AI tools and the cognitive processes activated during problem-solving; other supplementary studies investigate teachers’ beliefs, practices, and their capacity to integrate AI into instructional design and in the classroom. Drawing primarily on Brousseau’s Theory of Didactical Situations, the research employs mixed methods including questionnaires, semi-structured interviews, and comparative studies. Initial findings reveal that AI demonstrates inconsistent handling of semiotic representations, particularly visual ones, yet can function as a component of students’ adidactical environment - as part of the milieu - that promotes the emergence of mathematical concepts. Students appear to engage in a “double devolution dialectics”, shifting responsibility and commitment between themselves and AI in ways that can enhance mathematical autonomy. Teachers show various levels of critical engagement with AI outputs. The core study involves longitudinal observations of three teachers and approximately 100 students in authentic classroom settings.