Productive and Unproductive AI Usage in Programming Education: Effects on Learning Behaviour, Creativity, and Performance with the Moderating Role of AI Literacy and Mindfulness

<ns5:p>Background Artificial Intelligence (AI) technologies such as generative coding assistants and conversational AI systems are increasingly transforming programming education. While AI tools can enhance learning, creativity, and coding efficiency, excessive or uncritical dependency may undermine deep learning and independent problem-solving abilities. Despite growing adoption of AI in higher education, limited empirical research has differentiated between productive and unproductive AI usage patterns and their effects on programming-related educational outcomes, particularly within developing country contexts. Methods This study employed a quantitative cross-sectional research design using survey data collected from 350 undergraduate programming students enrolled in Information Technology and Engineering programmes at the Sri Lanka Institute of Advanced Technological Education (SLIATE). Data were analyzed using Structural Equation Modeling (SEM) with SmartPLS and SPSS. The study examined the effects of productive and unproductive AI usage on learning behaviour, creative programming behaviour, and programming performance, while also testing the moderating roles of AI literacy and mindfulness. Results The findings indicate that productive AI usage significantly enhances learning behaviour (β = 0.42, p &lt; 0.001), creative programming behaviour, and programming performance (β = 0.28, p &lt; 0.001). Conversely, unproductive AI usage negatively affects learning behaviour (β = −0.18, p &lt; 0.001) and programming performance (β = −0.22, p &lt; 0.001). Learning behaviour significantly predicts creative programming behaviour (β = 0.45, p &lt; 0.001), which subsequently improves programming performance (β = 0.39, p &lt; 0.001). AI literacy and mindfulness demonstrate significant moderating effects, strengthening productive AI engagement while reducing the harmful consequences of unproductive AI dependency. The structural model demonstrates substantial explanatory power, explaining 56% of the variance in programming performance. Conclusions The study demonstrates that AI functions as a double-edged educational technology in programming education. Productive AI usage supports self-regulated learning, creativity, and academic performance, whereas unproductive dependency undermines cognitive engagement and skill development. The findings emphasize the importance of AI literacy, mindfulness, and responsible AI integration within higher education curricula to ensure sustainable and effective AI-assisted learning environments.</ns5:p>