This paper analyzes the educational effects of a peer-learning agent based on pair programming in programming courses. A peer-learning agent system was developed to facilitate the learning of a programming language through the use of pair programming strategies. This system is based on the role of a peer-learning agent from pedagogical and technical aspects and simulates the "tutor" and "tutee." The peer-learning agent uses artificial intelligence methods with a Bayesian network as well as teaching and learning methods that simulate pair programming. This paper develops a model for determining students' programming abilities. In addition, the roles of the tutor and tutee are like the roles of a navigator and driver in pair programming. The developed agent system is demonstrated to have positive effects on knowledge retention and transfer in a programming course, with a greater influence on transfer than on retention. This model combining peer-learning agents with a teaching and learning strategy is more effective in helping learners to acquire programming skills. (Contains 8 figures and 4 tables.)