A Design-Oriented Framework for Mediating Generative AI in Programming Learning Technologies: Moderators, Causal Pathways, and Operational Guidelines

Generative Artificial Intelligence (GenAI) systems are increasingly integrated into programming education, yet empirical evidence reports highly heterogeneous learning outcomes, ranging from productivity gains to erosion of fundamental competencies. These inconsistencies suggest that learning effectiveness is not an intrinsic property of GenAI, but rather an emergent result of how learning technologies are designed, orchestrated, and embedded within instructional contexts. This study aims to derive design oriented knowledge to inform the development and evaluation of GenAI based learning technologies for programming education. We conducted a systematic mapping of 101 peer-reviewed studies published between 2020 and 2025 across seven digital libraries, following PRISMA 2020 guidelines. The synthesis integrates thematic analysis and comparative evidence to identify pedagogical mediation strategies, moderators of effectiveness, and process level transformations in programming learning supported by GenAI. The results reveal three interdependent design dimensions that shape learning outcomes in GenAI powered learning technologies: (i) temporal design, emphasizing phased introduction of AI following the consolidation of programming fundamentals; (ii) structural design, operationalized through calibrated scaffolding, adaptive assistance, and fade-out mechanisms; and (iii) instructional design, focusing on metacognitive prompting, reflective code evaluation, and process-oriented tasks. Across the corpus, three critical moderator interactions explain the prevalence of mixed outcomes: knowledge level × timing of AI introduction, meta-cognition × type of GenAI system, and mediation design × usage patterns. These interactions expose systematic risks such as dependency and illusion of competence, particularly when generic large language models are deployed without pedagogical control. Based on these findings, we propose a design oriented integrative framework that articulates inputs, mediation mechanisms, moderator interactions, and multidimensional learning outcomes through three empirically grounded causal pathways. The framework is accompanied by testable propositions and operational guidelines that translate empirical evidence into actionable design principles for GenAI based programming learning technologies. The study contributes actionable insights for researchers and practitioners seeking to design, implement, and evaluate learning technologies that leverage generative AI while preserving deep conceptual understanding and learner