Course-embedded AI-augmented teaching is associated with engagement and performance in software education

AI-augmented teaching may reduce feedback delays in software courses, but semester-long classroom evidence remains limited. This course-level quasi-experimental study evaluated a course-embedded AI-augmented teaching (AIT) model in an undergraduate Python course. Two intact classes participated: 60 students in the AIT class used a large language model teaching assistant restricted to course-specific materials and instructional rules for question answering, code diagnosis, personalized recommendations, and teacher monitoring, whereas 60 students in the conventional-instruction class received the same course instruction without access to the AIT platform. Student-level analyses, interpreted cautiously because of the class-level design, showed higher posttest, programming assignment and final composite scores in the AIT class after baseline adjustment. Learning engagement also showed clearer gains, especially overall and behavioral engagement. AI usage intensity was moderately associated with engagement but weakly associated with academic performance. Mediation analysis provided clearer evidence for an indirect engagement pathway for project practice performance than for final composite performance. Open-ended feedback indicated that students valued immediate feedback and debugging support but remained concerned about accuracy, dependency and ethical boundaries.