‘Resistance is futile?’: identity tensions and principled selectivity in AI-integrated teaching

Synthesis: Adiozaman and Segar (2026) interviewed two experienced academics three times across a single teaching semester and found that AI integration does not force a choice between embracing and resisting it. Instead it generates three recurring identity tensions — pedagogy versus platform, educator versus facilitator, and care versus compliance — which are navigated through what the authors call principled selectivity: context-sensitive decisions guided by pedagogical values, ethical commitment and professional judgment. The contribution is a conceptual lens that treats teachers' uneven AI use as judgment rather than as incomplete adoption.

Overview

Research on generative AI in higher education has concentrated heavily on adoption: who uses the tools, how often, under what conditions. Far less is known about what sustained AI-mediated teaching does to experienced academics' sense of themselves as teachers — their Teaching, their authority, their relationship to students. This study starts from that gap and asks how professional identity is negotiated when AI becomes a standing feature of the teaching environment rather than an experiment.

The authors take an interpretivist position: identity is treated as situated, evolving and socially constructed through practice, not as a fixed attribute a teacher either has or lacks. Two frameworks anchor that view. Wenger's communities of practice supply the idea that identity is constituted through participation and the negotiation of meaning, and phenomenological inquiry privileges first-person accounts of lived experience as the primary site of meaning-making. On this account, the interesting question is not whether a teacher has adapted, but how the disruptions AI introduces to professional practice are interpreted and worked through over time.

Study Design & Method

The design is a semester-long, prospective qualitative study that also incorporates retrospective reflection. Following the convention that longitudinal designs presuppose a span of at least a year, the authors are careful not to claim longitudinal status: their contribution lies in structuring reflection prospectively and retrospectively across one teaching semester.

• Participants. Two higher education academics, each with more than ten years of teaching experience and mixed teaching-research appointments, purposively selected from a larger parent sample as contrasting cases. Active engagement with generative AI was required — deliberate incorporation into learning tasks, Assessment design or feedback, or sustained pedagogical responses to students' independent AI use — while routine uses such as presentation software or a learning management system did not qualify. Engagement was established by self-report at screening and confirmed before the first interview.
• The two cases. Eva deliberately integrated AI into brainstorming, collaborative learning, formative feedback, curriculum design and digital communication within a broader learner-centered approach. Ace took a more selective route, using platforms for discussion and reflection while responding to students' AI use rather than building AI into tasks.
• Data. Three rounds of semi-structured interviews per participant across the semester, supported by reflective prompts administered at multiple points. The interview guide is reproduced as an appendix and covers teaching values, current and anticipated AI roles, felt tensions, and the kind of teacher each participant wanted to become.
• Analysis. Reflexive thematic analysis in five stages: familiarization with transcripts, inductive coding of segments related to identity and tensions, development of first-order concepts kept close to participants' language, abstraction into second-order themes, and synthesis into aggregate dimensions capturing identity work and negotiation. A data-structure table reports the first-order concepts, second-order themes and aggregate dimensions.

Key Findings

• Three interrelated tensions organize the data. They are pedagogy versus platform, educator versus facilitator, and care versus compliance — and they are interrelated rather than separate problems, since boundary-setting in one reshapes the others.
• Pedagogy versus platform runs from anticipation through disruption to ongoing negotiation. Both participants held a depth-oriented commitment in which learning is not supposed to be easy, and both saw AI's shortcut affordances as a threat to intellectual effort, which turned into monitoring of AI use and, later, active negotiation of its place.
• Role tension centers on displacement and expertise. The educator-versus-facilitator tension captured concerns about role displacement and the standing of expertise, surfacing through questions about authority and relevance rather than about tool competence.
• Care versus compliance is where ethics and workload meet. Authorship concerns, integrity and care for students clustered with overwhelm, "keeping up", and uncertainty, so the ethical and the practical were entangled: protecting students' intellectual work and protecting one's own capacity turned out to be the same conversation.
• Neither participant adopted or resisted wholesale. Both described deliberate, context-sensitive decisions about when, how and how much AI belonged in their teaching, guided consistently by pedagogical values, ethical considerations and professional judgment.
• Principled selectivity develops in stages across the semester. It begins as an implicit orientation — Ace framing AI as a complement rather than a substitute in order to preserve cognitive effort and independent thinking, Eva applying evaluative judgment by asking whether a tool actually deepens learning, widens access or strengthens connection. By mid-semester it is enacted: Ace asked students to explain their thinking and shifted emphasis to in-class discussion while holding boundaries around AI use, and Eva redesigned tasks and reassessed what students should learn, treating AI as something shaped by pedagogy rather than adopted. By semester's end it stabilizes without becoming fixed — Ace arriving at cautious acceptance that does not reject AI outright but stays intentional about not letting teaching become purely tool-driven, Eva describing her practice as still evolving and constantly adapting.
• Selectivity is an iterative three-part process. Across both accounts it moves through acceptance, adaptation and refusal, with refusal a legitimate move rather than a failure of uptake. Ace's remark that resistance is "quite futile" and that he therefore focused on working with AI meaningfully signals strategic engagement, not resignation.
• Identity is not resolved, it is continuously negotiated. Rather than stabilizing through alignment, identity emerges as a dynamic, relational process under sociotechnical change.

What this means for practice

• Instructors. Treat uneven AI use as a deliberate stance rather than a gap to close: decide task by task whether a tool deepens learning, widens access or strengthens connection, and be ready to explain that reasoning instead of apologizing for it.
• Instructors. When AI shortcuts threaten intellectual effort, redirect the task rather than police it — asking students to explain their thinking and moving emphasis to in-class discussion is what one participant arrived at by mid-semester.
• Faculty developers. Design development around judgment rather than tool fluency: prompt teachers to articulate what a tool changes about student thinking, access and connection before covering its features, and build repeated reflection points across a semester, because principled selectivity moved from an implicit orientation through enactment to a stabilized-but-evolving stance — an arc a single workshop cannot mirror.
• Administrators. Read selective non-use as professional judgment rather than lagging adoption; adding feature training addresses the wrong problem, while heavy workload and unsettled authorship policy push the care-versus-compliance tension toward compliance. Role-based tensions about authority and expertise are likewise not solved by tooling but by clarifying what the human contribution is meant to be in an AI-mediated course.
• Researchers. Study identity work over time rather than at a single point, since the tensions themselves shift — pedagogical tension intensified while role tension was reconfigured and ethical considerations consolidated into a sustained stance — and use the concept to describe teacher Learner Agency under technological change without implying that adaptation is the only competent response.

Limitations

• Two participants, both academics with more than ten years of teaching experience purposively selected as contrasting cases from a larger parent sample, so the findings are analytically rather than statistically generalizable.
• AI engagement was established by self-report at screening and confirmed before the first interview, and the data consist of three semi-structured interviews per participant plus reflective prompts rather than classroom observation or student outcome data.
• The design covers a single teaching semester, and the authors are explicit that this does not meet the year-long span they associate with longitudinal status.
• Identity work is situated in institutional, disciplinary and cultural contexts the study does not sample, and the authors call for work across wider ranges of participants and settings.

Citation

Adiozaman, I. F. A., & Segar, A. R. (2026). ‘Resistance is futile?’: identity tensions and principled selectivity in AI-integrated teaching. Teaching in Higher Education.