Аннотация
This study explored how pre-service physics teachers design artificial intelligence (AI)-based pedagogical infrastructure to support metacognitive learning in STEM classrooms. A qualitative interpretive design was employed with 30 pre-service physics teachers enrolled in a physics teacher education program at a university in Kazakhstan. Following two preparatory workshops, participants developed, tested, and refined AI-supported STEM lesson scenarios over 12 weeks. Testing involved peer-based micro-teaching with fellow pre-service teachers acting in simulated student roles; no school students participated. Data comprised individual semi-structured interviews, lesson plans, AI interaction scenarios, prompt sequences, and reflective notes and were examined using reflexive thematic analysis. Six interrelated themes were identified: designing AI-supported STEM lessons; supporting metacognitive regulation and formative assessment; aligning AI use with the physics curriculum and STEM objectives; preserving student agency and ensuring responsible AI use; developing teacher design agency; and addressing conditions for sustainable implementation. Participants' designs generally shifted from using AI to generate ready-made materials toward integrating it more purposefully into structured learning sequences. Pedagogically meaningful scenarios required learners to formulate an initial prediction or solution before consulting AI, verify AI-generated responses using physical laws, calculations, experiments, and reliable sources, and reflect on and revise their reasoning. Reported challenges included inaccurate AI responses, preparation time, unequal access to digital resources, technical difficulties, and variations in the quality of Kazakh- and Russian-language outputs. The findings highlight the importance of preparing pre-service physics teachers for subject-specific, ethically responsible, and pedagogically grounded AI integration. They illustrate design practices intended to scaffold planning, monitoring, scientific verification, evaluation, and reflection. However, the findings concern participants' pedagogical designs and professional perceptions and do not establish improvements in school students' metacognitive regulation or learning outcomes.