Paper deep dive

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

Jennie Ren, Jordan H. McDowell, Kyrie Zhixuan Zhou

Year: 2026Venue: arXiv preprintArea: cs.HCType: PreprintEmbeddings: 44

Intelligence

Status: not_run | Model: - | Prompt: - | Confidence: 0%

Entities (0)

No extracted entities yet.

Relation Signals (0)

No relation signals yet.

Cypher Suggestions (0)

No Cypher suggestions yet.

Abstract

Abstract:Generative AI is reshaping programming education, yet educators often infer students' AI-supported learning from classroom observations alone. This experience report presents a trio-ethnography involving two computing educators with different teaching philosophies and one undergraduate computer science student to examine how these interpretations evolve through dialogue. Across three conversations, the educators reflected on students' AI use, discussed changes to programming pedagogy, and revisited their assumptions after engaging with the student's lived experiences. Rather than simply confirming or contradicting the educators' perspectives, the student's narratives revealed learning processes that were largely invisible in the classroom, prompting both educators to reconsider assumptions about AI use, assessment, transparency, and programming instruction. We argue that trio-ethnography offers a valuable reflective approach for helping computing educators move beyond observable student behaviors toward a richer understanding of AI-supported learning and for informing instructional adaptation in the era of generative AI.

PDF

Open source PDF →Open local PDF →

Trouble viewing inline? Open PDF directly →

Full Text

42,532 characters extracted from source content.

Expand or collapse full text

Beyond Perspectives: A Trio-Ethnography of Interpretation Evolution in LLM-Supported Programming Education Jennie Ren ren_c@mercer.edu Mercer University United States Jordan H. McDowell Jordan.H.Mcdowell@live.mercer.edu Mercer University United States Kyrie Zhixuan Zhou kyrie.zhou@utsa.edu University of Texas at San Antonio United States Abstract Generative AI is reshaping programming education, yet educators often infer students’ AI-supported learning from classroom obser- vations alone. This experience report presents a trio-ethnography involving two computing educators with different teaching philoso- phies and one undergraduate computer science student to examine how these interpretations evolve through dialogue. Across three conversations, the educators reflected on students’ AI use, discussed changes to programming pedagogy, and revisited their assumptions after engaging with the student’s lived experiences. Rather than simply confirming or contradicting the educators’ perspectives, the student’s narratives revealed learning processes that were largely invisible in the classroom, prompting both educators to reconsider assumptions about AI use, assessment, transparency, and program- ming instruction. We argue that trio-ethnography offers a valuable reflective approach for helping computing educators move beyond observable student behaviors toward a richer understanding of AI- supported learning and for informing instructional adaptation in the era of generative AI. CCS Concepts • Social and professional topics→ Computing education. Keywords Computing education, Programming education, Generative AI, Large language models, Trio-ethnography 1 Introduction Generative AI has rapidly transformed programming education by changing how students learn programming, complete programming assignments, and solve programming problems. As large language models (LLMs) become increasingly integrated into programming learning, they can generate solutions to programming assignments while also explaining concepts, debugging errors, and supporting in- dependent learning. Consequently, observable programming prod- ucts, such as submitted code or completed assignments, no longer necessarily reflect students’ learning processes and outcomes.As a result, a growing gap has emerged between what educators can ob- serve from programming products and how students actually learn with AI, making it increasingly difficult for educators to accurately interpret students’ learning in the LLM era. This gap has important implications for programming educa- tion because effective instruction depends on educators’ ability to accurately interpret how students learn.Educators make decisions about assessment, feedback, classroom activities, and instructional strategies based on how they interpret students’ learning. When these interpretations are inaccurate or incomplete, instructional decisions may become misaligned with students’ actual learning needs, resulting in instructional practices that are less effective in supporting student learning. Therefore, there is a need for ap- proaches that help educators develop more accurate interpretations of students’ AI-supported learning. To address this need, we employed trio-ethnography involving two computing educators and one undergraduate computer science student.Rather than using trio-ethnography simply to compare ed- ucators’ and students’ perspectives on AI-supported programming learning, we used it as a process for developing more accurate in- terpretations of students’ learning through dialogue.Interestingly, we found that accurate interpretations did not emerge immedi- ately.Instead, educators’ interpretations evolved throughout the dialogue as their initial assumptions were refined, confirmed, or complicated through engagement with the student’s lived experi- ences.This interpretive evolution ultimately enabled educators to develop more accurate understandings of students’ AI-supported learning and reconstruct their teaching beliefs. This study makes both methodological and educational contri- butions. Methodologically, it demonstrates how trio-ethnography can be used not merely to compare educators’ and students’ per- spectives, but to examine how educators’ interpretations evolve through dialogue toward more accurate understandings of students’ AI-supported learning. By narrowing the gap between educators’ interpretations and students’ actual learning processes, this method- ological contribution enables an educational contribution by sup- porting the reconstruction of educators’ teaching beliefs based on more accurate interpretations of students’ learning. These recon- structed teaching beliefs provide implications for programming ed- ucation in the LLM era by informing pedagogical reflection and the design of instruction that better supports students’ AI-supported learning. 2 Related Work 2.1 Generative AI in Programming Education Recent computing education research has examined how genera- tive AI is reshaping programming learning, teaching, and assess- ment. Large language models can generate programming solutions, explanations, exercises [4,24], and responses to help-seeking re- quests [11], creating new opportunities for instructional support [9] while also raising concerns about over-reliance and academic in- tegrity [4]. Studies of novice programmers using tools such as ChatGPT and GitHub Copilot show that students use generative AI to generate code [14], understand syntax, debug errors, and seek arXiv:2607.22463v1 [cs.HC] 24 Jul 2026 Jennie Ren, Jordan H. McDowell, and Kyrie Zhixuan Zhou explanations [22], and complete programming tasks [23]. Also high- lighted are risks for novice learning, including reduced metacog- nitive engagement and difficulty evaluating generated code [23], overconfidence, and uneven benefits across learners [20]. Other work has examined how instructors respond to generative AI in computing education. Researchers have studied instructor perceptions [21,27,29], course policies [2,15], AI-assisted assess- ment [12], AI-based tutoring systems [28], prompt-based program- ming exercises [8], and emerging guidelines for responsible class- room integration [6,17,19]. Collectively, this literature shows that generative AI is not merely a code-generation tool but a technology that reshapes programming practice, learning support, assessment design, and classroom policy. However, most studies character- ize AI tools, student uses, instructor perceptions, or educational risks. Less attention has been paid to how educators develop more accurate interpretations of students’ AI-supported learning, espe- cially when observable programming products may not reveal the learning processes behind them. 2.2 Dialogic Approaches to Understanding Educators and Students Prior research has used qualitative and dialogic approaches to understand educational experiences, identities, and perspectives. Duoethnography positions participants as co-inquirers who place their lived experiences in dialogue to question assumptions and generate new understandings [25, 26]. In education, duoethnogra- phy and related reflective approaches have been used to examine professional identity [5,18], language teacher education [3], and col- laborative reflection [1,7]. Trio-ethnography extends this dialogic tradition by bringing three participants into reflective conversa- tion and has been used in HCI, accessibility [13], and education research [10,16] to surface multiple lived experiences around tech- nology, disability, professional practice, and learning. These studies demonstrate the value of dialogic methods for surfacing multiple perspectives and supporting reflective under- standing. However, dialogue is primarily used to understand or compare participants’ perspectives, such as what educators think, what students experience, and how these perspectives differ. Our study uses trio-ethnography for a different purpose by positioning student dialogue as evidence that helps educators refine and recon- struct their interpretations of students’ AI-supported learning. In this way, trio-ethnography becomes a mechanism for narrowing the gap between what educators can observe and how students actually learn with AI. 3 Method 3.1 Research Design To investigate how trio-ethnography can help narrow the gap be- tween what educators can observe and how students actually learn with AI, this study employed this approach to examine how com- puting educators’ interpretations of students’ AI-supported pro- gramming evolved through dialogue and subsequently reshaped their teaching beliefs and instructional practices. Trio-ethnography positions participants as co-inquirers who critically examine, ques- tion, and reconstruct one another’s interpretations. Because gen- erative AI is rapidly reshaping computing education, we selected trio-ethnography to capture not only how interpretations evolve through dialogue, but also how this process informs educators’ teaching beliefs and future instructional decision-making. 3.2 Participants The trio consisted of two computing educators (R1 and R2) and one undergraduate computer science student (R3). R1 is an assistant professor in Computer Science with experience teaching under- graduate courses, including programming, web development, Java, and mobile application development. R1 initially expressed signifi- cant concern about students’ over-reliance on AI-generated code. R2 is an assistant professor in Information Systems who teaches introductory programming to undergraduate business students and has adopted an open approach toward integrating generative AI into programming instruction. R3 is an undergraduate Computer Science student who regularly uses generative AI while learning programming. The student joined the dialogue as an equal contribu- tor whose lived experiences extended and challenged the educators’ interpretations. The trio was intentionally composed of two edu- cators with contrasting instructional perspectives and one student who regularly used generative AI, enabling educators to examine and reflect on their interpretations through direct engagement with the student’s perspectives. 3.3 Procedure The study was conducted in three analytical stages. Stage 1: Initial Educator Dialogue. The two computing educators first engaged in a series of duo-ethnographic discussions guided by three reflection topics: (1) how generative AI has changed students’ programming learning, (2) how computing educators should adapt programming instruction in response to these changes, and (3) how these changes have influenced their teaching beliefs. Across these discussions, the educators articulated and developed their initial interpretations of students’ AI-supported learning based on their classroom experiences. As they questioned and reflected on one another’s perspectives, some of these interpretations were further refined through the dialogue itself. Stage 2: Student Dialogue. An undergraduate computer science student was subsequently interviewed using a semi-structured pro- tocol designed to explore the student’s lived experiences of learning programming with generative AI. Instead of mirroring the educa- tors’ discussion prompts, the interview questions were developed to elicit student perspectives that could respond to, elaborate on, and challenge the educators’ initial interpretations. The student’s narratives confirmed, extended, and complicated the educators’ initial interpretations of AI-supported learning. Stage 3: Educator Reflective Reconstruction. Following the student dialogue, the educators revisited their original discussions, treating the student’s dialogue as evidence to refine and recon- struct their interpretations of students’ AI-supported learning. This reflective process reconstructed the educators’ interpretations of students’ AI use, enabling more accurate understandings of stu- dents’ AI-supported learning, refined their teaching beliefs, and informed subsequent instructional practices. Beyond Perspectives: A Trio-Ethnography of Interpretation Evolution in LLM-Supported Programming Education 3.4 Data Analysis The conversations from Stage 1 (Initial Educator Dialogue) and Stage 2 (Student Dialogue) served as the primary data sources for analysis. These conversations were audio-recorded, transcribed using Otter.ai, and manually reviewed for accuracy. Through multiple rounds of transcript reading and collaborative discussion, we examined how educators’ interpretations of students’ AI-supported programming learning were confirmed, extended, challenged, or complicated by the student’s lived experiences. We also paid particular attention to recurring educator interpretations, as these reflected deeply held assumptions and often became the most meaningful sites of interpretive reconstruction. Rather than viewing these interactions as isolated observations, we treated them as evidence of an evolving interpretive process. By comparing the educators’ initial interpretations with the stu- dent’s perspectives, we reflected on how these interpretations were reconstructed through dialogue. This reconstruction subsequently refined the educators’ teaching beliefs and informed future instruc- tional practices in programming education. 4 Results Following data analysis, the findings reveal an evolving interpre- tive process through which educators gained increasingly accurate understandings of students’ AI-supported programming learning through dialogue with the undergraduate student. The findings il- lustrate how the student’s lived experiences prompted educators to re-examine their initial interpretations and reshape their teaching beliefs. 4.1 From AI Stigma to AI Usage Transparency to Explicit AI Guidance The educators’ interpretations of students’ AI use evolved from identifying an emerging AI stigma, to recognizing the importance of AI usage transparency, and ultimately to emphasizing explicit guidance for appropriate AI use. These interpretations, initially developed through educator dialogue, were later confirmed by the student’s experiences and reconstructed into new teaching beliefs. The dialogue began with R1 describing students who submitted programming assignments containing syntax beyond what had been taught in class while claiming they had not used AI. “Some students used syntax that I never taught in class... but in their reflection they still said they didn’t use AI.” Initially, R1 interpreted this behavior as students deliberately concealing AI use because they believed it was academically inap- propriate. “I think the students feel shameful about using AI... probably they think the instructor does not allow us to use AI.” Rather than viewing nondisclosure simply as dishonesty, the edu- cators interpreted it as evidence of an emerging AI stigma, in which students perceived AI use as something hidden from instructors. The dialogue then shifted toward instructors’ responsibility. R2 questioned whether expectations about AI use had ever been ex- plicitly communicated, prompting R1 to reflect: “I didn’t say anything. I didn’t teach them what should be a good way to use AI, or what should be a bad way to use AI.” Through this reflection, the educators’ interpretation shifted from viewing AI stigma as a student issue to recognizing that it was also shaped by instructional ambiguity. Rather than assuming students understood classroom expectations, they recognized that instructors needed to communicate those expectations explicitly. This refined interpretation led R1 to conclude: “I think AI usage transparency is very important.” Here, AI usage transparency extended beyond encouraging dis- closure. Instead, it meant making instructors’ expectations visible so students understood when AI use was appropriate and how it could support learning. Comparing their teaching practices further refined this interpre- tation. Unlike R1, R2 intentionally discussed AI expectations during the first class meeting. “I spent maybe ten minutes in the first class telling students what is allowed and what is not allowed with AI use.” This comparison led both educators to recognize that trans- parency alone was insufficient. Students also needed explicit guid- ance on how AI should support programming learning. Their inter- pretation therefore evolved beyond transparency toward explicit AI guidance, emphasizing that instructors should teach produc- tive AI-supported learning strategies rather than simply establish classroom rules. The subsequent student dialogue confirmed this refinement. The student described feeling uncertain about AI use because AI- assisted programming had been prohibited in high school. The student also expressed a preference for instructors to communicate expectations explicitly and, more importantly, teach students how AI could be used effectively for learning rather than simply whether it was permitted. Reflecting across both educator and student dialogues, the edu- cators reconstructed their understanding of AI-supported program- ming instruction. Their reconstructed teaching beliefs emphasized that AI should become an explicit, discussable, and teachable compo- nent of programming education through transparent expectations and purposeful guidance that enables students to use AI to support, rather than replace, their learning. 4.2Redefining AI’s Role: From Answer Provider to Learning Partner and Tutor A second interpretive thread centered on how the educators un- derstood AI’s educational role in programming learning. Although both educators initially expressed concern that students were treat- ing AI as an answer provider that replaced independent thinking, their dialogue refined this interpretation by recognizing that AI itself was not the educational problem. Instead, the educators came to view AI as a learning partner, and more specifically, as a tu- tor that could scaffold students’ conceptual understanding while preserving meaningful learning. This refined interpretation was Jennie Ren, Jordan H. McDowell, and Kyrie Zhixuan Zhou subsequently confirmed by the student’s lived experiences and rein- forced through educator reflection, positioning AI as a pedagogical learning partner in programming education. Initially, R1’s interpretation was shaped by observations that many students relied on AI to obtain programming solutions rather than understand programming concepts. “I feel like the students just got used to copy and paste the code... they’re not using AI to help them, they’re just using AI like, ‘I can get the answer from AI,’ and just paste the answer.” Rather than criticizing AI itself, R1’s concern centered on its educational role. When AI functioned as an answer bank, students risked replacing their own reasoning with AI-generated solutions, reducing opportunities for conceptual learning. The dialogue gradually refined this interpretation. Rather than debating whether AI should be permitted, R2 encouraged recon- sidering how AI could support learning. Drawing on software en- gineering practice and his teaching experience, he argued that AI- generated code was not inherently problematic; students’ learning depended on how they engaged with it. “I don’t really care if students use AI to generate the initial version of the code... if they have prior expe- rience with programming by themselves, I don’t feel like AI coding is a major concern.” This discussion shifted the educators’ interpretation from view- ing AI as an answer provider toward recognizing it as a learning partner. More specifically, AI was increasingly conceptualized as a tutor that helps students understand programming concepts, ex- plain unfamiliar syntax, and support problem solving while leaving learning in students’ hands. The student dialogue strongly confirmed this refined interpre- tation. Rather than describing AI as a tool for obtaining complete solutions, the student consistently portrayed it as a learning re- source for understanding Java syntax, programming concepts, and code. “It was really helping me out... I was studying the syntax... Now you can learn the syntax.” Throughout the interview, the student described interacting with AI by asking questions, seeking explanations, and improving conceptual understanding rather than simply requesting complete answers. These narratives aligned with the educators’ refined in- terpretation that AI could function as a tutor supporting students’ learning. Reflecting across both educator and student dialogues, the ed- ucators reconstructed AI’s educational role. Their reconstructed teaching beliefs emphasized positioning AI as a learning partner and tutor that supports conceptual understanding, reasoning, and independent learning. 4.3 Invisible Learning Beyond AI-Generated Answers A third interpretive thread centered on educators’ persistent inter- pretation that students often copied AI-generated answers without meaningful learning. Unlike the previous themes, this interpretation was complicated rather than confirmed by the student’s experiences. During the educator dialogue, R1 refined this concern through the idea of “one more step”: AI could support learning if students con- tinued thinking after receiving AI-generated code. The student later complicated this refined interpretation by revealing multiple learn- ing activities that occurred after AI-generated responses—activities that often remained invisible to instructors. Initially, R1 interpreted students’ AI use as obtaining answers rather than learning programming. From this perspective, AI-generated code appeared to replace students’ own reasoning. “I feel like the student, they just got used to copy paste the code... they’re not using the AI to help them, they’re just using the AI, like, oh, I can get the answer from the AI, and just paste the answer.” This interpretation was grounded in classroom evidence. R1 could observe students’ submitted code, including syntax or pro- gramming techniques beyond what had been taught. However, these observable products provided limited access to what students had actually done before submission. As a result, AI-generated or unusually advanced code was often interpreted as evidence of copy-and-paste behavior rather than meaningful learning. As the educator dialogue unfolded, this interpretation was re- fined. R2 suggested that AI-generated code was not inherently problematic if students continued engaging with it. R1 therefore reframed the issue as whether students took an additional step after receiving AI-generated solutions. “If they have a further step... they use the code gener- ated by AI, then they can step on this kind of thing to have the further thinking, that’s fine, but if they just... copy paste the code, that’s not fine.” This idea of “one more step” became a turning point in the edu- cators’ interpretation. The educators distinguished between AI use that ended with answer generation and AI use that initiated fur- ther reasoning. However, they still had limited evidence regarding whether students actually engaged in these further learning steps. The subsequent student dialogue complicated this refined inter- pretation. The student confirmed that AI could reduce the effort required to obtain a solution, but also described additional learning activities after receiving AI-generated help. “Usually I have notes to take down, I practice, and I will just create more of what I was trying to solve the first time, and just make sure I’m able to do this from muscle memory.” The student’s narrative expanded the educators’ idea of “one more step” into multiple learning steps. After receiving AI-generated assistance, the student described reading code, analyzing its logic, taking notes, practicing independently, modifying solutions, and verifying understanding. These activities suggested that AI interac- tion did not always end with copying a final answer; AI-generated responses often became the beginning of a longer learning process that remained largely invisible to instructors. These accounts complicated the educators’ persistent interpreta- tion without fully invalidating it. Some students may indeed copy AI-generated code without reflection. However, the student narra- tives revealed that similar programming submissions could result from different learning trajectories. One student might submit code Beyond Perspectives: A Trio-Ethnography of Interpretation Evolution in LLM-Supported Programming Education after copying an AI-generated answer, while another might submit a similar product after reading, questioning, modifying, practicing, and verifying understanding. From the instructor’s perspective, both submissions may appear similar, but the learning processes behind them may differ substantially. Reflecting across both educator and student dialogues, the edu- cators reconstructed their understanding of AI-supported learning. Their reconstructed teaching beliefs emphasized recognizing the invisible learning processes that may occur beyond AI-generated answers. 4.4 Reconstructing Teaching Beliefs Through Student Dialogue The final interpretive thread centered on how student dialogue reshaped the educators’ teaching beliefs in the LLM era. As a result, the educators began to reimagine assessment, debugging instruc- tion, and active learning opportunities. The student dialogue revealed reasoning processes that required greater instructional support in AI-supported programming learn- ing. Although the student described AI as helpful for learning syntax and programming concepts, the dialogue also revealed situations where AI produced a working solution without helping the student understand why it worked. When comparing personal code with AI-generated code, the student explained that the two versions appeared similar, yet only the AI-generated version worked. “What I had versus what the AI had... it was pretty similar, but mine was incorrect. While the AI’s was correct, I wasn’t really sure why and how that was different.” This student perspective prompted the educators to reconsider what programming instruction needed to make visible. The educa- tors recognized the importance of helping students reason about why code worked, why errors occurred, and how AI-generated solutions differed from their own attempts. The student dialogue reshaped the educators’ beliefs about de- bugging instruction. The student’s difficulty in understanding why AI-generated code worked while personal code failed suggested that students needed explicit support in comparing code, interpret- ing errors, and reasoning through debugging processes. During the educator dialogue, R1 similarly reflected that debugging instruction should focus not only on fixing errors but also on understanding the reasoning behind errors. “Instead of only focusing on fixing errors... they also need to understand the source, the reason behind the errors.” This reconstructed teaching belief positioned AI as a tool to sup- port students’ debugging reasoning by helping them explain errors, compare alternative solutions, and understand why corrections work. The educators also reconsidered assessment. When students could use AI to quickly generate correct code, completed assign- ments no longer necessarily represented students’ learning out- comes. This prompted the educators to reconsider assessment de- signs that evaluate students’ reasoning in addition to final code, such as monitored assessments or assignments requiring students to explain, justify, or reflect on their solutions. Such opportunities could also make the invisible learning processes identified in Theme 3 more visible to educators. Finally, the educators’ reconstructed teaching beliefs empha- sized active learning opportunities that reveal students’ reasoning. Across the dialogue, both educators increasingly valued practices such as live coding, active recall, and code-output prediction. R1 described future instruction as creating more opportunities for stu- dents to actively learn rather than passively follow teacher-guided demonstrations. “As a teacher, I mean to, in the future, I mean to design more this kind of opportunities for the students, so they can learn.” Reflecting across both educator and student dialogues, the edu- cators reconstructed their teaching beliefs to emphasize instruction that reveals, supports, and develops students’ reasoning. These re- constructed teaching beliefs emphasized programming instruction that assesses reasoning, makes learning processes visible, teaches debugging explicitly, and creates active learning opportunities that support learning with AI. 5 Discussion 5.1 Implications for Programming Education Across the trio-ethnographic dialogue, educators’ interpretations of students’ AI-supported programming learning evolved rather than remained fixed. Student dialogue confirmed key educator interpre- tations, such as AI transparency and AI as a learning partner, while also complicating persistent assumptions by revealing learning processes that were largely invisible from classroom observations. These findings shift the focus from judging whether students use AI to understanding how they learn with AI. Observable evi- dence, such as submitted code or completed assignments, provides only partial access to students’ reasoning and learning processes. Incorporating students’ perspectives therefore enables educators to develop more accurate understanding of AI-supported program- ming learning. More importantly, the trio-ethnographic dialogue extended be- yond understanding students to reshaping educators’ teaching be- liefs. These reconstructed teaching beliefs, in turn, prompted edu- cators to reconsider their programming instruction. The following implications illustrate how this reconsideration informed broader pedagogical reflection in the LLM era. 5.1.1 Repositioning AI in Programming Education. A key impli- cation of this study is the need to reposition AI in programming education from an issue of classroom policy to a component of pedagogy. The reconstructed teaching beliefs point toward two comple- mentary pedagogical directions. First, AI should become a visible, legitimate, and teachable part of programming classrooms. Edu- cators should explicitly communicate when AI use is appropriate, discuss responsible AI use, and provide guidance on how AI can ef- fectively support learning. Second, programming instruction should position AI as a learning partner and tutor that supports students’ conceptual understanding, problem-solving skills, and independent reasoning. Jennie Ren, Jordan H. McDowell, and Kyrie Zhixuan Zhou Collectively, these implications shift programming education from regulating AI use to designing learning environments where students learn with AI. 5.1.2 Making Learning Processes Visible. A third implication con- cerns making students’ AI-supported learning processes more vis- ible. The findings showed that instructors may observe students’ final code or assignment submissions, but these products provide limited evidence of what students actually did after receiving AI- generated assistance. Meaningful learning may occur through read- ing, analyzing, practicing, modifying, and verifying AI-generated responses, yet these processes often remain invisible in the final programming product. For programming education, instructors should design activities that make students’ reasoning and learning processes more visible. Reflection prompts and process documentation can ask students to explain how they used AI, what they accepted or rejected, what they modified, and what they learned. Such practices shift attention from detecting AI use to under- standing how students learn with AI, allowing educators to better distinguish between copy-and-paste submission and meaningful AI-supported learning. 5.1.3 Methodological Implications of Trio-ethnography. Beyond its implications for programming education, this study highlights the methodological value of trio-ethnography as a way to examine how educational interpretations evolve through dialogue, rather than merely collecting different perspectives from educators and a student. This suggests that trio-ethnography moves beyond comparing perspectives by using student dialogue as evidence to progressively refine educators’ interpretations and develop more accurate under- standings of student learning. This methodological contribution ultimately informs pedagogical reflection and programming educa- tion in the LLM era. 5.2 Limitations and Future Work This study has several limitations that should be considered when interpreting the findings. First, the trio-ethnography involved only one undergraduate stu- dent participant. Although the student’s perspectives provided rich insights into AI-supported programming learning, the participant represented a highly motivated learner. Consequently, the student learning strategies described throughout the dialogue, such as read- ing AI-generated code, taking notes, practicing independently, and verifying understanding, may not represent the experiences of all programming students. Future research should therefore include students with more diverse academic backgrounds, learning mo- tivations, and programming abilities to examine whether similar interpretive patterns emerge across different learner populations. The study also focused on a single trio-ethnographic dialogue involving two programming educators and one undergraduate student. The reconstructed interpretations and teaching beliefs should therefore be understood as contextually situated rather than universally representative. Future studies could conduct trio- ethnographies across different institutions, programming courses, and educational contexts to examine how educators’ interpretations evolve in diverse learning environments. 6 Conclusion This study demonstrates the methodological value of trio-ethnography in computing education. By using student dialogue as evidence to progressively refine educators’ interpretations, trio-ethnography supported more accurate understandings of students’ AI-supported programming learning. These reconstructed understandings reshaped educators’ teach- ing beliefs and informed pedagogical reflection for programming education in the LLM era. We hope this work encourages future computing education research to further explore the methodologi- cal potential of trio-ethnography. 7 Ethics and Privacy Statement This experience report presents a trio-ethnographic inquiry among the authors. All conversations involving the educators’ and stu- dent’s reflections were conducted voluntarily with informed con- sent. To protect privacy, pseudonyms (R1, R2, and R3) are used through- out the paper. The findings reflect the participants’ individual expe- riences rather than universal practices and are intended to support transparent, responsible, and equitable integration of generative AI in computing education. Beyond Perspectives: A Trio-Ethnography of Interpretation Evolution in LLM-Supported Programming Education References [1]Tony E Adams, Stacy Holman Jones, and Carolyn Ellis. 2016. Handbook of autoethnography. Routledge. [2]Areej Ali, Aayushi Hingle Collier, Umama Dewan, Nora McDonald, and Aditya Johri. 2025. Analysis of generative AI policies in computing course syllabi. In Proceedings of the 56th ACM Technical Symposium on Computer Science Education V. 1. 18–24. [3]Dario Luis Banegas and David Gerlach. 2021. Critical language teacher education: A duoethnography of teacher educators’ identities and agency. System 98 (2021), 102474. [4]Brett A Becker, Paul Denny, James Finnie-Ansley, Andrew Luxton-Reilly, James Prather, and Eddie Antonio Santos. 2023. Programming is hard-or at least it used to be: Educational opportunities and challenges of ai code generation. In Proceedings of the 54th ACM Technical Symposium on Computer Science Education V. 1. 500–506. [5]Yasemin Cakcak Tezgiden and Ufuk Ataş. 2024. Becoming and being a critical language teacher educator: A duoethnography. (2024). [6]Doga Cambaz and Xiaoling Zhang. 2024. Use of ai-driven code generation models in teaching and learning programming: a systematic literature review. In Proceedings of the 55th ACM Technical Symposium on Computer Science Education V. 1. 172–178. [7] Heewon Chang, Faith Ngunjiri, and Kathy-Ann C Hernandez. 2016. Collaborative autoethnography. Routledge. [8]Paul Denny, Juho Leinonen, James Prather, Andrew Luxton-Reilly, Thezyrie Amarouche, Brett A Becker, and Brent N Reeves. 2024. Prompt Problems: A new programming exercise for the generative AI era. In Proceedings of the 55th ACM Technical Symposium on Computer Science Education V. 1. 296–302. [9]Paul Denny, James Prather, Brett A Becker, James Finnie-Ansley, Arto Hellas, Juho Leinonen, Andrew Luxton-Reilly, Brent N Reeves, Eddie Antonio Santos, and Sami Sarsa. 2024. Computing education in the era of generative AI. Commun. ACM 67, 2 (2024), 56–67. [10] Smit Desai, Tanusree Sharma, and Pratyasha Saha. 2023. Using ChatGPT in HCI research—A trioethnography. In Proceedings of the 5th International Conference on Conversational User Interfaces. 1–6. [11]Arto Hellas, Juho Leinonen, Sami Sarsa, Charles Koutcheme, Lilja Kujanpä, and Juha Sorva. 2023. Exploring the responses of large language models to beginner programmers’ help requests. In Proceedings of the 2023 ACM Conference on International Computing Education Research-Volume 1. 93–105. [12] Muntasir Hoq, Yang Shi, Juho Leinonen, Damilola Babalola, Collin Lynch, Thomas Price, and Bita Akram. 2024. Detecting ChatGPT-generated code sub- missions in a CS1 course using machine learning models. In Proceedings of the 55th ACM Technical Symposium on Computer Science Education V. 1. 526–532. [13] Dhruv Jain, Venkatesh Potluri, and Ather Sharif. 2020. Navigating graduate school with a disability. In Proceedings of the 22nd International ACM SIGACCESS Conference on Computers and Accessibility. 1–11. [14]Majeed Kazemitabaar, Justin Chow, Carl Ka To Ma, Barbara J Ericson, David Weintrop, and Tovi Grossman. 2023. Studying the effect of AI code generators on supporting novice learners in introductory programming. In Proceedings of the 2023 CHI conference on human factors in computing systems. 1–23. [15] Sam Lau and Philip Guo. 2023. From" Ban it till we understand it" to" Resistance is futile": How university programming instructors plan to adapt as more students use AI code generation and explanation tools such as ChatGPT and GitHub Copilot. In Proceedings of the 2023 ACM Conference on International Computing Education Research-Volume 1. 106–121. [16]Giang Nguyen H Le, Vuong Tran, and Trang Thuy Le. 2021. Combining photog- raphy and duoethnography for creating a trioethnography approach to reflect upon educational issues amidst the COVID-19 global pandemic. International Journal of Qualitative Methods 20 (2021), 16094069211034391. [17]Rongxin Liu, Carter Zenke, Charlie Liu, Andrew Holmes, Patrick Thornton, and David J Malan. 2024. Teaching CS50 with AI: leveraging generative artificial intelligence in computer science education. In Proceedings of the 55th ACM technical symposium on computer science education V. 1. 750–756. [18] Amy E Long, Rachel Wolkenhauer, and Mary Higgins. 2021. Learning through Research in a Professional Development School: A Duoethnographic Approach to Teacher Educator Professional Learning. School-university partnerships 14, 1 (2021), 36–44. [19]Joseph Maguire, Rosanne English, Qi Cao, and Chee Kiat Seow. 2025. Themes in the declared use of generative artificial intelligence in assessment. In Proceedings of the 9th Conference on Computing Education Practice. 17–20. [20]Lauren E Margulieux, James Prather, Brent N Reeves, Brett A Becker, Gozde Cetin Uzun, Dastyni Loksa, Juho Leinonen, and Paul Denny. 2024. Self-regulation, self-efficacy, and fear of failure interactions with how novices use LLMs to solve programming problems. In Proceedings of the 2024 on Innovation and Technology in Computer Science Education V. 1. 276–282. [21] James Prather, Juho Leinonen, Natalie Kiesler, Jamie Gorson Benario, Sam Lau, Stephen MacNeil, Narges Norouzi, Simone Opel, Virginia Pettit, Leo Porter, et al.2024. How instructors incorporate generative AI into teaching computing. In Proceedings of the 2024 on Innovation and Technology in Computer Science Education V. 2. 771–772. [22]James Prather, Brent N Reeves, Paul Denny, Brett A Becker, Juho Leinonen, Andrew Luxton-Reilly, Garrett Powell, James Finnie-Ansley, and Eddie Antonio Santos. 2023. “It’s weird that it knows what i want”: Usability and interactions with copilot for novice programmers. ACM transactions on computer-human interaction 31, 1 (2023), 1–31. [23]James Prather, Brent N Reeves, Juho Leinonen, Stephen MacNeil, Arisoa S Randri- anasolo, Brett A Becker, Bailey Kimmel, Jared Wright, and Ben Briggs. 2024. The widening gap: The benefits and harms of generative ai for novice programmers. In Proceedings of the 2024 ACM Conference on International Computing Education Research-Volume 1. 469–486. [24] Sami Sarsa, Paul Denny, Arto Hellas, and Juho Leinonen. 2022. Automatic generation of programming exercises and code explanations using large language models. In Proceedings of the 2022 ACM Conference on International Computing Education Research-Volume 1. 27–43. [25]Richard D Sawyer and Darren Lund. 2012. Duoethnography: Dialogic methods for social, health, and educational research. Vol. 7. Left Coast Press. [26]Richard D Sawyer and Joe Norris. 2013. Duoethnography: Understanding quali- tative research. [27] Judy Sheard, Paul Denny, Arto Hellas, Juho Leinonen, Lauri Malmi, and Simon. 2024. Instructor perceptions of ai code generation tools-a multi-institutional interview study. In Proceedings of the 55th ACM Technical Symposium on Computer Science Education V. 1. 1223–1229. [28]JD Zamfirescu-Pereira, Laryn Qi, Björn Hartmann, John DeNero, and Narges Norouzi. 2025. 61a bot report: Ai assistants in cs1 save students homework time and reduce demands on staff.(now what?). In Proceedings of the 56th ACM Technical Symposium on Computer Science Education V. 1. 1309–1315. [29]Cynthia Zastudil, Magdalena Rogalska, Christine Kapp, Jennifer Vaughn, and Stephen MacNeil. 2023. Generative ai in computing education: Perspectives of students and instructors. In 2023 IEEE Frontiers in Education Conference (FIE). IEEE, 1–9.

Instant research discovery