Multiple studies report that Computer Science (CS) instructors face problems on how to integrate visualizations in their teaching. This problem gets compounded for instructors in technology-constrained classrooms that are common in developing countries. In these classrooms, students are not able to interact with visualization directly; instead, their interaction is mediated by the instructor who alone may have access to the visualization. In the current study, we contrasted learning outcome from integrating program visualization at two different engagement levels in instructor-mediated classroom setting. The two levels were "Responding" (prediction activity with visualization) and "Viewing" (watching visualization with instructor commentary) as per Naps' taxonomy. The study was conducted for a programming topic of medium complexity. We found the strategy of prediction with visualization ("Responding") led to statistically significant higher active behavioral engagement and higher perception of learning among students than the strategy of watching the visualization with instructor commentary ("Viewing"). We also found statistically significant higher cognitive achievement in terms of the rate of problem solving for the "Responding" group, if the students had prior training in active learning. This study can serve as a reference guide to design effective integration of visualizations in instructor-mediated classrooms.