This study investigates the impact of AI-powered virtual assistants (e.g., ChatGPT, educational chatbots) on teachers’ pedagogical decision-making. Employing a sequential exploratory mixed-methods design (QUAL → QUAN), the research involved in-depthinterviews and focus groups with 16 educators, followed by a survey of 412 secondary and higher education teachers. Findings reveal that AI does not replace but reconfigures teacher decision-making into a hybrid cycle of diagnosis, AI consultation, and critical evaluation. The core conclusion is that teacher agency the capacity for critical judgment and contextual adaptation is the strongest predictor of positive outcomes. While AI enhances personalization and efficiency, its benefits are significantly moderated by institutional support and unresolved ethical concerns regarding bias, transparency, and learning authenticity. The study contributes an evidence-based, agency-mediated model of technology adoption, highlighting that the ultimate impact of AI in education depends less on algorithmic power and more on empowered professionals who can critically integrate these tools into their practice.