Abstract:

As generative AI embeds itself in academic work, educators must continually decide which AI outputs to trust, challenge or reject. This one‑year longitudinal self‑study examines how a teacher educator's sense of agency and emotional engagement evolved across 52 purposively sampled interactions with generative AI, selected to maximise variation in AI role, task complexity and emotional intensity. These interactions spanned four distinct roles, including co-author, data analyst, editor and counsellor. The interactions were selected in order to capture variations in task complexity and emotional intensity.

Drawing on Priestley et al.'s ecological model of teacher agency and research on epistemic emotions (curiosity, confusion, relief and unease), the study treats agency not as a fixed attribute but as something constantly renegotiated in response to AI outputs and the emotions they provoke. Data were captured through an innovative and bespoke AI–Agency Interaction Log (AAIL), recording contextual detail, perceived stakes, decision outcomes and emotional shifts across each episode.

The study illuminates the dynamics of educator agency in human-AI collaboration, showing that agency does not develop linearly towards either full automation or categorical rejection. Instead, three recurring cycles are identified: delegation and relief, discomfort and contestation, and re-entry and reassertion. These cycles are patterned by the role AI assumes: when AI functions as a data analyst, the educator experiences heightened dependence and reduced confidence, whereas AI as co-author tends to prompt intensified critical scrutiny. Throughout, epistemic emotions play a central mediating role, shaping the educator’s willingness to accept, modify or reject AI-generated output. The study offers two key contributions with direct implications for teacher education and AI policy. First, it advances a role-differentiated, emotion-attuned framework for understanding educator agency in ongoing collaboration with generative AI. Second, it introduces the AI-Agency Interaction Log (AAIL), a replicable tool for capturing the moment-to-moment dynamics of human-AI interaction that conventional survey or interview designs struggle to access.

Keywords:

Generative artificial intelligence, teacher agency, epistemic emotions, human–AI collaboration, teacher education, autoethnography, qualitative methodology.