Configurational pathways of programming learning beliefs and self-efficacy among college students using intelligent agents: evidence from GenAI literacy and agent usage experience

With the rapid penetration of Generative Artificial Intelligence (GenAI) in higher education, college students’ psychological dispositions, such as learning beliefs and self-efficacy, are being profoundly influenced during programming learning. Using a C++ programming course, this study investigates how multiple condition combinations shape students’ programming learning beliefs and self-efficacy in GenAI-supported environments. Based on questionnaire data collected from 394 college students after they engaged in programming tasks using an AI-Agent developed by the research team, this study employs fsQCA to identify configurational pathways to high levels of these dispositions, covering system quality, human-computer interaction, and learner attributes. The findings reveal that high programming learning beliefs and programming self-efficacy can be achieved through different combinations of technological, human–AI interaction, and learner-related conditions. System quality, information quality, human–AI trust, perceived value, and usage experience jointly contribute to favorable outcomes, with roles varying across configurations. Notably, GenAI-related knowledge and computational thinking play compensatory or substitutive roles across pathways, reflectingcomplementarity between external technological support and internal learner capabilities. This study advances a Human–AI Cognitive Symbiosis framework, conceptualizing learning beliefs and self-efficacy as emergent outcomes of configurational alignment among technological affordances, cognitive mediation, and psychological regulation, offering theoretical and practical insights for GenAI-supported programming instruction.