When Generative AI is used primarily for content generation in education, it risks undermining teacher agency by removing crucial opportunities for professional growth and creative thinking. This presentation examines theoretical and practical considerations for developing an intermediate application layer between teachers and Foundation Models, while also using Malderez's (2015; 2024) Systematic Informed Reflective Practice (SIRP) protocol as an exemplar framework for preserving teacher agency in AI interactions. Drawing on Lepage & Collin's (2024) warning that "AI systems can shift the centre of expertise from teachers and school administrators to programmers or system designers," we argue that both direct AI interaction and that which relies heavily on content generation cannot reach the upper levels of Septiani, Kostakos, & Romero's (2023) framework of creative engagement. To approach "Level 6: Expansive learning supported by AI," an intermediate application layer is essential and, moreover, must be developed alongside pedagogical experts while taking a critical stance on content generation; any content which is generated avoids undermining teacher agency by restricting it to that which the teacher chooses to delegate to AI. We illustrate the similarities between SIRP and structured AI prompting through a detailed mapping of mentor practices to AI capabilities and best practises. This mapping shows how SIRP's step-by-step protocol aligns naturally with ‘chain-of-thought’ and ‘few-shot’ AI prompting, and how its emphasis on mentor listening mirrors AI's capacity for ‘infinite patience’ in supporting reflection. Moreover, the ONSIDE framework (Hobson, 2016) provides clear role definitions that can be effectively encoded in prompts, ensuring interactions remain at least guided by good mentoring principles. We then show that a combination of visible AI dialogue and invisible background processes can suitably model various aspects of mentor thinking, especially when further combined with ‘knowledge’ containing e.g. curricula, teaching standards, teaching materials and other contextual information. This approach positions AI as a scaffolding tool that helps teachers draw their own conclusions on the basis of their experience (Malderez, 2009), thereby preserving their role as experts in their professional growth. The presentation concludes by proposing a checklist for evaluating the goodness-of-fit of existing practices for AI adaptation, derived from the aforementioned mapping. This checklist provides a practical framework for institutions seeking to implement AI-supported teacher development systems while safeguarding teacher agency and autonomy. We will finish with some examples of how systematically working through this checklist provides a new lens through which to revisit, revise and improve existing mentor practices.

Hobson, A. (2016). Judgementoring and how to avert it: introducing ONSIDE Mentoring for beginning teachers. International Journal of Mentoring and Coaching in Education, 87-110. Lepage, A., & Collin, S. (2024). Preserving Teacher and Student Agency: Insights from a Literature Review. In A. Urmeneta, & M. Romero, Creative Applications of Artificial Intelligence in Education (pp. 17-34). Cham: Palgrave Studies in Creativity and Culture. Malderez, A. (2009). Mentoring. In A. Burns, & J. C. Richards, The Cambridge Guide to Second Language Teacher Education (pp. 259-268). New York: Cambridge University Press. Malderez, A. (2015). On Mentoring in Supporting (English) Teacher Learning: where are we now? Inspirations in Foreign Language Teaching – Studies in language pedagogy and applied linguistics. Malderez, A. (2024). Mentoring Teachers: Supporting Learning, Wellbeing and Retention. Routledge. Septiani, D., Kostakos, P., & Romero, M. (2023). Analysis of creative engagement in AI tools in education based on the #PPai6 framework.