Generative AI as a learning companion: opportunities and challenges for novice programmers

Generative artificial intelligence (GenAI) is increasingly utilized to support programming education, yet its educational value for novice programmers remains uncertain because existing research is fragmented across pedagogical, technical, and ethical domains. This study aimed to synthesize current evidence on the opportunities and challenges of using GenAI as a learning companion in novice programming education and to derive implications for responsible instructional integration. A structured literature review was conducted using peer-reviewed records indexed in Scopus and published between 2023 and 2025. A Boolean search combined terms related to GenAI, learning-support functions, novice learners, and programming contexts. Scopus AI outputs, including summaries, a concept map, topic-author mapping, and emerging themes, were used as discovery and clustering aids, while the final analysis traced claims to the cited publications and compared convergent and conflicting findings. The review found that GenAI can provide personalized explanations, adaptive scaffolding, immediate feedback, coding examples, and scalable support, particularly in large classes or settings with limited instructor availability. However, the literature also identified substantial risks, including inaccurate or biased outputs, academic dishonesty, privacy concerns, unequal access, and over-reliance that may weaken debugging, problem solving, and independent reasoning. The findings therefore indicate that GenAI is most defensible as a mediated support tool rather than an autonomous tutor. Its educational value depends on explicit usage boundaries, AI literacy, source verification, transparent disclosure, and assessment designs that make student reasoning visible. The study implies that institutions should adopt pedagogy-led governance, provide educator development, and sequence AI support so that scaffolding can be withdrawn as competence develops. Because the review used one database and did not include classroom experiments, interviews, or control groups, future research should use multi-database, longitudinal, and controlled designs to test whether GenAI support produces durable independent programming competence.