AI multi-agent collaboration in instructional design: a quasi-experimental study on need satisfaction, AI literacy, and performance

Instructional design is essential for pre-service teachers, but its development is hindered by limited mentoring and fragmented feedback. This study developed an AI-based multi-agent system to scaffold multi-level instructional design. A quasi-experimental design was employed with an AI-supported group and a teacher-led group . Results indicated that teacher-led guidance significantly enhanced autonomy , competence , and relatedness satisfaction, whereas no comparable improvements were observed in the AI-supported group. Across affective, behavioural, and cognitive dimensions, no significant effects emerged for AI literacy. While overall instructional design performance did not differ significantly between groups, rubric-based analyses revealed that the teacher-led group achieved greater gains in critical areas such as analysing learners’ prior knowledge, assessing breadth and depth, and organising content.AI multi-agent systems offer structured, scalable support but cannot replace human guidance for motivational and higher-order pedagogical needs.