<![CDATA[The rapid adoption of Generative Artificial Intelligence (GenAI) in higher education has shifted research attention from whether students use AI to how they use it meaningfully. However, research examining how students build academic context before interacting with AI remains limited, with most studies focusing on prompt engineering as a technical strategy for improving AI output quality. This study explores how a pre-service mathematics teacher demonstrates learner agency when interacting with GenAI in a Real Analysis course. Employing a single qualitative case study design, data were collected from one purposively selected participant through screen recordings, learning presentations, and semi-structured interviews. Data analysis followed a case study approach incorporating pattern matching against two theoretical frameworks: the Three AIEd Paradigms comprising AI-directed (learner-as-recipient), AI-supported (learner-as-collaborator), and AI-empowered (learner-as-leader) and Mathematics Noticing, encompassing the dimensions of attending, interpreting, and deciding. Findings reveal four interconnected patterns of learner agency: establishing academic common ground before AI interaction, critically evaluating AI responses through multi-source verification, detecting conceptual gaps in AI explanations, and positioning AI as a confirmation tool for independent thinking. These patterns reflect a developmental progression from AI-supported toward AI-empowered positioning within a single interaction session, with the three noticing dimensions developing in an integrated and reinforcing manner throughout the process. This study concludes that learner agency is the critical factor enabling productive AI use in mathematics learning not the frequency or accuracy of AI responses, but students' capacity to critically manage their interactions with AI.]]>
