The rapid integration of GenAI into teacher education programs has created a critical imperative to explore its emerging impact on how PSTs develop professional skills and make pedagogical judgements. This paper reports on a pedagogical innovation that uses Cogniti AI as a reflective partner in a classroom research course designed for preservice English teachers. The study examined the tensions generated through PSTs navigating two distinct modes of engagement: a "confirmatory" mode oriented toward validation and reassurance, and an "exploratory" mode oriented toward generating alternatives and broadening perspectives. By positioning AI as a mediating tool, this study analysed how PSTs moved from passive recipients to active collaborators who exercised transformative agency—the capacity to reconceptualise existing frames of action to adopt new strategies to transform their classroom practice.

Research Questions

1. How do preservice teachers use Cogniti AI tool to develop their classroom-based research strategies?

2. How does this tool mediate preservice teachers' agency in the development of their pedagogical/research decisions used in designing classroom-based research?

Research Objectives

• Analyse how PSTs utilise AI tools to critically examine the coherence between their research focus and the development of classroom-based evidence.

• Understand how AI-mediated interactions shape PSTs’ sense of professional agency in research and pedagogical decision-making.

• Explore how Cogniti AI can be most effectively used as a pedagogical tool to support preservice teachers' classroom-based research.

Theoretical Framework

Given this research focus, the study is conceptually framed by Cultural-Historical Activity Theory (CHAT), which understands learning as a culturally and historically situated process mediated by social and material tools. From a CHAT perspective, human agency is understood not as an innate trait but as an intentional, deliberate set of actions taken to transform social reality.

Further, central to this research is the related conception of transformative agency, which aligns with the drive of expansive learning central to CHAT theorising (Engeström, 1987). This orientation proposes that in confronting contradictions between the current and desired state, actors can move beyond existing knowledge and practices to create new expansive actions. Consistent with this conception, GenAI is understood in this study as a mediating artefact—that is, not merely a tool that facilitates a task—but instead something that fundamentally disrupts the mental processes and professional development of PSTs. 

In addition, we used positioning theory to understand the discursive shifts that occur during this mediation. As PSTs interact with AI tools, they necessarily need to negotiate their professional authority, in essence, "positioning" themselves as either dependent users seeking validation or as empowered researchers (i.e., using the GenAI as a sounding board to solve complex pedagogical problems). Through the CHAT lens, we sought to understand PST interactions with GenAI as a site of professional transformation where agency is fostered through the resolution of conflicts between current practices and new affordances emerging through GenAI.

This research adopts a qualitative, interpretive case study design to capture the nuances of PSTs’ interactions with AI tools. The participants were preservice English teachers enrolled in a university-based teacher education program who engaged in classroom-based research tasks supported by Cogniti AI. To ensure a deep and triangulated understanding of their experiences, data were collected from multiple sources over a semester-long intervention period. The primary data source was AI Conversation Logs, which provided comprehensive insights into the interactions between PSTs and the AI. These logs provided an objective basis for understanding how students iterated their prompts and the depth of analytical questioning adopted. This was complemented by two other data sources—Student Reports and Reflective Journals—where PSTs gave meaning to their subjective experiences, capturing their emerging engagement with the affordances of Cogniti AI. Finally, a series of focus group interviews explored the collective tensions and shifts in pedagogical roles that occurred when AI was introduced into the design of the classroom-based research process. Data analysis followed a multi-layered approach, using thematic analysis to identify patterns of engagement. This centred on identifying evidence of changes in agency and expansive learning actions—such as questioning, analysing, and modelling—to distinguish between stances of dependence and empowerment. Furthermore, positioning analysis was used to analyse how PSTs and teacher educators discursively constructed their authority and roles whilst using Cogniti AI in decision-making processes.

The findings demonstrate a progressive reconfiguration of PSTs’ professional agency in their engagement with Cogniti AI, unfolding across four interrelated stages: embryonic, emerging, evolving, and expanded. In the embryonic stage, PSTs engaged with the tool in largely confirmatory ways, focusing on technical prompt refinement, accuracy checks, and validation of pre-existing ideas. This mode of engagement played an important pedagogical role, providing psychological safety and reducing uncertainty in early research work, yet agency remained primarily procedural. As PSTs progressed, their engagement shifted through emerging and evolving stages toward more exploratory and reflective uses of Cogniti AI. Students increasingly positioned it as a reflective partner, asking it to confirm coherence between research questions, instruments, and evidence, and to surface tensions or alternative analytic possibilities. In these stages, Cogniti AI functioned as a mediator of metacognition, prompting PSTs to interrogate the adequacy of their methodological choices and to exercise professional judgement by selectively adopting, adapting, or rejecting AI feedback. At the expanded stage, agency was expressed most clearly in PSTs’ capacity to decide when and how GenAI should be used—or deliberately not used—based on pedagogical, ethical, and developmental considerations. Across all stages, a persistent tension was the risk of a dependency trap, whereby the fluency and ease of GenAI output could undermine independent professional judgement. Learning to navigate this tension, rather than maximising GenAI use, emerged as central to the development of professional agency. The theoretical contribution of this research is to conceptualise GenAI as a relational mediator in the development of teacher agency. Pedagogically, it offers guidance on scaffolding GenAI use to foster critical autonomy rather than dependence. Practically, it provides design principles for research tasks in which GenAI supports professional judgement and disciplined reflection, enhancing—rather than replacing—human teaching and inquiry.

Alé, J., Ávalos, B., & Araya, R. (2025). Chilean teachers’ knowledge of and experience with artificial intelligence as a pedagogical tool. Education Sciences, 15(10), 1268. https://doi.org/10.3390/educsci15101268 Edwards, A. (2015). Working relationally in and across practices: A cultural-historical approach to collaboration. Cambridge University Press. Engeström, Y. (1987). Learning by expanding: An activity theoretical approach to development research. Helsinki: Orienta-Konsultit. Harré, R., & van Langenhove, L. (1999). Positioning theory: Moral contexts of intentional action. Blackwell Publishers. Martin, J. (2020). Teacher agency and the sociotechnical: Reconfiguring professional learning through relational approaches. Teaching and Teacher Education, 92, 103046. https://doi.org/10.1016/j.tate.2020.103046 Lodge, J. M., Yang, S., Furze, L., & Dawson, P. (2023). It’s not like a calculator, so what is the relationship between learners and generative artificial intelligence? Learning: Research and Practice, 9(2), 117–124. https://doi.org/10.1080/23735082.2023.2261106 Reyes-Rojas, J., Díaz, B., Ruz-Reveco, C., Castro, A., & Reyes-González, D. (2026). Conceptualizing pre-service teachers’ readiness for AI integration into teaching practices: An intelligent-TPACK approach. Computers and Education Open, 10, 100320. https://doi.org/https://doi.org/10.1016/j.caeo.2025.100320