This study examines Socratic Mind, a generative artificial intelligence (GenAI)-powered tool designed to scale adaptive Socratic dialogue and provide reasoning-focused practice that is difficult to deliver at scale in large courses. Implemented in an undergraduate computer science course, the study adopts a mixed-methods approach that integrates academic outcomes, behavioral indicators of tool use, sentiment analysis of student-AI dialogue transcripts, and students’ reflective responses about higher-order thinking. Behavioral traces and dialogue features suggest that productive engagement was characterized by sustained, elaborated interaction, including persistence across assignments and richer dialogic exchanges. Affective analyses indicate that students expressed a wider range of emotional tones during dialogue in AI-mediated learning. Students’ reflections further suggest that adaptive Socratic questioning supported cognitive engagement, especially among fully engaged students, by prompting them to articulate reasoning, confront misconceptions, and monitor their understanding. Overall, our findings provide actionable insight for designing pedagogically meaningful and engaging GenAI-assisted educational tools.