The integration of artificial intelligence (AI) tools in higher education has transformed how students engage with academic tasks, particularly in project-based learning environments. The research investigates the impact of AI-assisted learning on student self-efficacy within an Information Systems program at a higher education institution in Malang, Indonesia. An experimental approach was implemented in two courses, Testing and Documentation and Digital Transformation, where students utilized AI tools, including ChatGPT, Google Gemini, and Consensus, to develop scientific reports based on real-world projects. Data were collected through two stages of Focus Group Discussions (FGDs) involving 20 students selected through random sampling, supported by open-ended interview questions and qualitative analysis. The findings reveal a dynamic relationship between AI usage and self-efficacy. At the initial stage, students reported increased confidence due to AI’s ability to support idea generation, structuring, and writing efficiency. This led to a temporary decline in self-efficacy, highlighting students’ limited ability to critically evaluate AI-generated content. In the second stage, following structured guidance from the lecturer on validation techniques and critical review, students demonstrated improved confidence and stronger analytical skills. Their self-efficacy increased as they gained control over the process of verifying AI outputs, resulting in more scientifically valid and accountable reports. This study contributes to the literature by demonstrating that self-efficacy is not a static outcome of AI adoption but a dynamic construct influenced by instructional design. The results emphasize the importance of lecturer-led scaffolding in transforming AI from a dependency tool into a catalyst for critical thinking. The findings provide practical implications for designing AI-integrated learning strategies that balance technological support with the development of student autonomy and academic integrity.