Work

Artificial intelligence in programming education: A systematic review of intelligent, adaptive, and generative systems, learner outcomes, and implementation challenges

Year: 2026

Type: review

Abstract: This systematic review synthesises empirical evidence on learner-facing artificial intelligence (AI) in programming education. It examines how intelligent, adaptive, generative, recommender, assessment, and feedback systems support programming-skill acquisition; what motivational, engagement, collaborative, and problem-solving outcomes are reported; and what ethical, technological, accessibility, and access-related challenges arise. Following PRISMA guidelines, a Scopus search identified English-language journal articles published between 2010 and 2025, of which 76 studies met the narrowed eligibility criteria. The findings suggest that AI-supported systems appear most educationally valuable when they combine personalised learning pathways, learner-facing assessment, adaptive scaffolding, and timely process-oriented feedback. Reported outcomes include improvements in programming performance, conceptual understanding, motivation, engagement, self-efficacy, collaboration, and problem-solving, although effects vary according to learner experience, prior knowledge, task type, and instructional design. Motivational benefits were reported across adaptive, conversational, reflective, collaborative, problem-based, and immersive environments; however, explicitly game-based or gamified evidence was limited. The most consistently documented implementation challenges concerned inaccurate or misleading AI outputs, overreliance, academic-integrity uncertainty, unequal access, and differences in learners' capacity to verify generated solutions. Direct empirical evidence on algorithmic bias, formal transparency mechanisms, privacy-preserving design, and institutional data governance remained sparse. Overall, AI should be understood as a complement to sound pedagogy, human judgement, peer interaction, and learner agency. Future research should use stronger comparative and longitudinal designs, evaluate independent learning after AI support is withdrawn, and examine reliability, equity, privacy, and subgroup differences more directly.

Source: Heliyon

DOI: 10.1016/j.heliyon.2026.e45361

Authors: Jozsef Katona, Klara Ida Katonane Gyonyoru

Institutions: Obuda University, Wekerle Business School, John von Neumann University, University of Dunaújváros

Language: English

Topic: Teaching and Learning Programming

Subfield: Computer Science Applications

Field: Computer Science

Domain: Physical Sciences

SDG: Quality education

Open Access status: gold

Estimated APC paid: $2,270