Human-AI Collaborative Learning Ecosystem: Effects of Multi-Agent-Based Programming Learning on Learning Outcomes, Computational Thinking, and Behavioral Patterns in Higher Education

Programming education plays a central role in computer science learning, and intelligent tutoring systems offer possibilities for enhancing instructional effectiveness. However, conventional programming instruction often faces challenges in providing support tailored to diverse cognitive processes and learner needs. This study proposes a Multi-Agent-based Programming Learning (MA-PL) approach, which leverages collaborative agent technology to provide differentiated support including algorithmic guidance, code assistance, and learning monitoring based on learners’ cognitive states and problem-solving progress. An experimental design was adopted with 54 university students randomly assigned to experimental and control groups for a 12-week intervention. Results indicated that the experimental group achieved significant improvements in learning outcomes and computational thinking, and exhibited more positive programming behavioral patterns characterized by higher-level cognitive engagement. However, the study revealed reduced autonomous construction behaviors and a shift from peer collaboration to human-AI collaboration. These findings offer empirical insights into multi-agent technology in programming education.