Beyond Perspectives: A Trio-Ethnography of Interpretation Evolution in LLM-Supported Programming Education

Synthesis: This experience report introduces trio-ethnography — structured dialogue between two computing educators with differing teaching philosophies and one undergraduate CS student — as a method for surfacing how educators' interpretations of students' AI use evolve. The central finding is that much AI-supported learning is invisible from the classroom: across three conversations, the student's lived-experience narratives revealed learning processes that neither educator could infer from observable behavior, prompting both to revise assumptions about AI use, assessment design, and transparency in CS Education.

The paper complicates the dominant framing of Student Experience research, which typically measures behavior (prompt logs, usage frequency) rather than interpretation. Educators here initially read student AI use through the lens of Over-Reliance risk, but dialogue revealed more nuanced self-regulation and Scaffolding-seeking than surveillance-style observation suggested. This aligns with broader calls for AI Literacy frameworks that treat students as reflective agents rather than compliance subjects.

Methodologically, trio-ethnography offers a low-cost reflective instrument for faculty development: rather than instrumenting classrooms with analytics, it uses sustained cross-perspective conversation to update pedagogical beliefs. As an experience report with n=3 it makes no generalizability claims, but it provides a replicable protocol for departments adapting programming instruction in the Generative AI era.

What this means for practice

• Instructors. State AI expectations explicitly in the first class meeting instead of assuming students infer them; one educator spent about ten minutes on what is and is not allowed, and the dialogue showed that AI stigma was partly produced by instructional ambiguity rather than by student choice.
• Instructors. Run a low-cost cross-perspective dialogue with one student before instrumenting a course with analytics; three structured conversations surfaced learning processes such as reading AI-generated code, taking notes, practicing independently, and verifying understanding that submitted code never revealed, which makes this a practical Educational Development instrument.
• Learners. Volunteer your reasoning, not just your artifacts: the student's lived-experience narratives were what complicated the educators' initial over-reliance reading of AI use.
• Instructors. Reposition AI from classroom policy to pedagogy by teaching when use is appropriate and how AI can support conceptual understanding, problem solving, and independent reasoning rather than only regulating it.

Limitations

• n = 3: two computing educators and one undergraduate student across three conversations, presented as an experience report that makes no generalizability claims.
• The single student participant was a highly motivated learner, so the reported strategies may not represent students with other academic backgrounds, motivations, or programming abilities.
• The dialogue was a single trio ethnography at one institution, so the reconstructed interpretations and teaching beliefs are contextually situated rather than universally representative.
• The inquiry was conducted among the authors themselves, with pseudonyms used to protect privacy, so the reconstructed interpretations reflect the researchers' own perspectives.

Citation

Ren, McDowell & Zhou (2026). Beyond Perspectives: A Trio-Ethnography of Interpretation Evolution in LLM-Supported Programming Education. arXiv preprint.