A week of fika: ACM ICER 2026
By the end of this academic year, I needed a vacation. It was a year of major disruptions to my federal research funding, weekly disruptions in my administrative role from the U.S. federal government, and weekly disruptions to my life from federal and state civil rights loss from anti-trans legislation. When I really thought about the kind of vacation I needed, I knew I needed a few things: 1) to leave the country while I can, 2) to reconnect with people around the world, 3) and to frolic with food and adventure. Eight days in Sweden, with ACM ICER 2026 conference in the middle, felt like a good fit. And the Swedish tradition of fika, a mid-day break for coffee, connection, and calm, felt like the right metaphor for my mid-summer mood.
Of course, ICER wasn’t isolated from many of those broader disruptions in the U.S. Conference registration was down heavily due to U.S. federal funding cuts. The mood amongst Ph.D. students was somber, with jobs and growth in the community stunted by research, immigration, and generative AI disruptions. And the broader future of the field, with generative AI also breaking CS curricula, assessment, and broader public sentiment about CS learning and learning in general, left me feeling like the community was flailing. And yet, the community is nothing if not plucky, kind, and curious, and so it was a cozy place to ponder what comes next together.
Note that I decided to keep this report short. This was a week of rest, and that included me staying present. Check out the open proceedings on the ACM Digital Library for details on any of the summaries below.
Wednesday wondering
I started Wednesday with a robust buffet breakfast and a sharp havre cortado.
Session: Secondary Learning
The first session of the day was a medley of talks about secondary education. My wonderful former student Jayne Everson (Foundry10) kicked off the first session talking about (the surprisingly rare) joyful student learning experiences, uncovering the importance of care and agency. Jesse Hoobergs (KU Leuven) talked about disconnection between learning and interest development, reinforcing the central role of motivation and identity in shaping future learning. Franz Jetzinger (TU Munich) shared some teacher experiences learning and teaching AI literacy in secondary settings in Germany, sharing some of the difficulty of evaluating real-world AI use, what counts as AI, and what constitutes an “AI” problem.
While the papers really had very little to do with each other explicitly, I found the underlying thread of lack of teacher support to be the most important theme, reinforcing how little respect and resources teachers are given to create vibrant learning experiences for youth. I’m biased, but Jayne’s work was the one that felt the most visionary for shaping futures that address these gaps, pointing to how much teachers can do with support, training, and freedom.
Session: Measurement
The next session considered how we know what students know. Jan Vahrenhold (University of Münster) talked about how cognitive load is associated with code complexity measures, finding some predictive power, but mostly source lines of code, and nothing rivaling expert teacher judgement. Next, James Prather (Abilene Christian University) shared a concise instrument for measuring self-efficacy using generative AI in programming, designed for diagnostic use in a teaching setting. Enrico Benedetti (Utrecht University) ended the session with a talk on computational thinking, problem solving, and post-secondary “Bebras” tasks in (problems that elicit computational thinking), discovering that at least in an introductory Python class, planning was most associated with correct answers, but that planning was hard. (No big surprise there: this is a robust finding from many pior works, in and beyond computing education, and has never been more important in a world of AI-assisted programming).
Across the first two papers, I had the same question: why? So much of good measurement design is about validation for a particular use, but the uses weren’t that clear to me. That poses a big risk of misapplication of these instruments, and over-interpretation of their meaning. Measurement is about precision, and we need more precision in how we conceptualize what instruments are for.
Session: Tools
After a tasty lunch, Brent Reeves (Abilene Christian University) talked about an approach to better scaffold student use of generative AI code suggestions, finding that offering multiple suggestions provided some productive friction, helping them think critically about their choice. Max Barth (ETH Zurich) presented a way of embedding AI programming tutors with self-regulated learning and cognitive engagement scaffolds, finding that such scaffolds didn’t change time on task or constructive contributions, but did foster longer engagement.
I found both of these tool visions a bit dystopia: they locked students into an interaction with an LLM-based model that had no awareness of their emotional states, their motivations, their goals. It’s no wonder that some students found the experiences frustrating. Intelligent tutoring systems have been reasonably predictable in their behavior up until now, potentially aggravating these experiences.
Session: Literature Reviews
After a vibrant poster break of doctoral consortium students and conversation, there were a two papers considering the field. Rita Garcia (Victoria University of Wellington) examined the prevalence of replication studies, finding that they are somewhat more prevalent, but scholars in our community are still anxious about doing them for all of the predictable reasons. The second paper, presented by Adaiti Allen Kadams (University of Eastern Finland) reviewed work on cognitive offloading and generative AI in the specific heterogeneous contexts of computing education in Africa, from an access, readiness, and relevance lens. There were only 11 papers in their review, but this eclectic group of studies set in African countries found evidence of cognitive offloading without some kind of reasoning expectation, and a strong need for multilingual, culturally relevant tools and content.
While I’m not a always a fan of how many literature reviews our community writes, such reviews are important: they remind us of where we’ve been, what we’ve learned. I’m glad these two were published.
Session: Lightning Talks
Immediately after the last session was a string of lightning talks. They covered reflective learning in generative AI, generative AI integration in data science education, post-secondary assessment redesign in light of generative AI, peer instruction tools, summer internship learning, responsible computing pedagogies, students’ generative AI reflections, and primary AI literacy pedagogy.
Did you notice any trends in the topics above? Folks are understandably chasing questions about generative AI, because its the learning technology of the moment, and driving hiring decisions in academia. But I was pleased to see many other equally important questions about other aspects of learning.
Thursday thoughts
To my dismay, some of the best coffee in Uppsala didn’t open until 1030, so I repeated my Thursday breakfast and mediocre cortado.
Session: Talking to synthetic text extruders
The morning started with Paul Denny (University of Auckland) sharing a study on student perceptions and strategies when initially describing computational tasks to LLMs to generate programs, finding many omissions of expected output, required inputs, and other basic requirements of problem specification, and that students would often not read and interpret the code generated. Solomon Oyelere (University of Exeter) extended this conversion, considering how to assess reasoning and fluency in students’ knowledge in light of their student generative AI use, especially in low power, unstable internet settings in Nigeria. Through a series of interviews, they found that subjectivity, device usage norms, and infrastructure were an amplifier of assessment threats, and offered an alternative model for scaffolding assessment around these threats. Victor-Alexandru Padurean (Max Planck Institute) presented the third paper on how students respond to (intentionally and unintentionally) defective AI-generated code, finding that students re-prompted more to “natural” defects, but edited more after injected defects, suggesting that injecting defects did create small practice opportunities.
These three papers represented a robust inquiry into the shifting dynamics of learning about programming when models generate mediocre code, and students do not yet have skills to prompt or evaluate effectively.
Session: Beyond programming
After a savory fika, we had a session on aspects of learning that went beyond programming. Keith Tran (NC State University) spoke on the effects of generative AI on student motivation through a social expentancy value theory lens, finding that students in their interviews shared a range of identity, utility, and value-based motivations for learning, shaping their perspectives on generative AI dependency. After, Adrienne Gifford (Issaquah Middle School) and I talked about an exploration of creative writing and creative coding integration with Wordplay, finding that computing changed what kinds of language arts learning was possible and language arts changed what kind of computer science learning was possible, creating a kind of mutually supporting learning through expression. The third talk from Anna Sollazzo (University of Glasgow) also explored transdisciplinary learning, but in digital humanities, finding computational literacy, problem solving, sociopolitical critiques, and critical, comparative pedagogies.
I’m biased, of course, but I thought this was the most interesting session so far, digging into the deeper emotional and identity factors that structure learning, and the more imaginative potential futures of computing education.
Session: Representation
After a cozy lunch break, Naaz Sibia (University of Toronto Mississauga) shared a study how students used multiple visualizations via think aloud, reflection, and gaze tracking, finding that they only considered code and memory, but not visual metaphors. Annapurna Vadaparty (University of California, San Diego) talked about the role of representation in planning and problem decomposition, exploring hierarchical diagrams and sequence representation (a la UML), direct instruction on decomposition. They found a variety of diagram authoring practices, errors in program quality and encapsulation, and many mixed notation practices similar to these in prior work on professional developers’ sketches of architecture. They speculated that students may be invoking incompatible notional machines.
Get Amy J. Ko’s stories in your inbox
Join Medium for free to get updates from this writer.
Both of these papers were interesting in the way that they wrestled with program understanding in non-code media, hinting at the many new priorities around reasoning about software implementation in abstract, opaque ways through generative AI.
Session: Accessibility
After a hybrid poster session, I chaired a session on accessibility. Jinyoung Hur and Michael Kang (University of Illinois Urbana-Champaign) examined major selection by students with disabilities from the CRA data buddies project, finding that students with disabilities were more likely to switch out of computing than students without disabilities, mostly due to poor inclusion, irrelevant curricula, and isolatio. Students with disabilities were more likely to stay because someone encouraged them to. Next, Chris Perdriau (also University of Illinois at Urbana-Champaign) spoke about Deaf student experiences with live coding in CS lectures, noting numerous multimodal and logistical access barriers, many of which could be addressed by writing live coding process steps down. The last talk by Alyson Yin (University of California, Irvine) surveyed student experiences in AccessComputing as a long-running intervention, finding that the most valued activities was community, conference funding, and mentorship.
I was so excited to see these three works (and three others) on accessibility and disability. It’s been such a gap in computing education research, and it feels like there’s a new critical mass of attention on the topic.
Castle time
After the last session was a quirky evening in Uppsala Castle. We learned Swedish traditional songs, heard live music, had conversation and were entertained by a mentalist. It was very… specific. I feel like I’d been welcomed to someone’s secret club, to live briefly on their terms, and in trade, they give me food, drink, and entertainment of their choice. I enjoyed much of it, was perplexed by some of it, and enjoyed being present with others processing their experiences.
Friday frictions
I had my same breakfast, but adventured to a new cafe and found myself a better cortado. I also had a fun time practicing my nascent Swedish and had the barista chuckle at my pronunciation.
Session: Introductory Something
The morning session was about introductory programming classes, or whatever introductory might become. Alex Chao (University of California, San Diego) shared a study on the role of teaching assistants relative to generative AI, framed around sociocultural theory. Through a set of interviews, he found that some students fundamentally valued human interaction, care, and community, and the social contracts around learning that people help uphold, and felt guilt and skepticism relying on generative AI. The next study from Daniel Manesh (Virginia Tech) returned to the prior session’s live coding, examining instructor practices, finding many practical theories about applications to teaching new concepts, reinforcing integrative practice, or modeling process. In practice, instructors tended to do a mix of all of these, and many nuanced intuitions about tradeoffs in engagement and code sharing. Kristin Fasiang (Northwestern University) presented the third paper, examining student practices around “siding” (back-channeling and side conversations) during direction instruction such as lectures. She found that tasks were often on task, and supplementing instructor focus; others were off task, but very much connected to relationship building and peer socio-emotional support.
I was particularly fond of Kristin’s paper in this session, since it was such a granular look at the social dimensions of learning. Alex’s talk was also a nice glimpse into the experiences of students with generative AI relative to teaching assistants.
Session: Pedagogy, Pairs, and Programs
After a break, Sverrir Thorgeirsson (ETH Zurich) discussed a machine learning method for synthesizing learning progressions, generating instructional sequences to minimize cognitive load on the topic of learning diffusion models. Their algorithm’s sequence led to better learning outcomes, though effect size was modest. Next, Nick Gardella (University of Virginia) experimented with variations on human pair programming, and human-generative AI pair programming. He found that the AI condition led to better on performance, but found no evidence of better learning; students suggested that they had better understanding when working with a person, and detrimental effects on students’ affect. Finally, Claus Brabrand (IT University of Copenhagen) discussed a study of why recursion is hard in the context of Use-Modify-Create scaffolding; students were a bit less correct with recursion and a bit slower, adding to the evidence that taking recursion after iteration seems to be hard.
Many of the discussions we had about these three talks centered on questions about who was in charge of deciding pedagogy: a machine, a teacher, students themslves? Many of these considerations seemed to be implicit assumptions in these works, and we wondered how that might have changed our interpretation of their results.
Session: Learning Data Structures and Algorithms with Billion Parameter Convolutional Neural Networks
After a cozy group photo outside, Quinton Yong (University of Victoria) talked about effects of data structure learning, comparing GenAI, visualization, and human supports; live tutor support resulted in the lowest increase in self-efficacy, but GenAI resulted in lower quiz scores, suggesting risks of GenAI producing overconfidence. The second in the short session was by Erica Goodwin and Katherine Braught (University of Illinois) on generating personalized and contextualized algorithm design problems with LLMs. Some examples had realism and authenticity, others did not.
This brief session showed that LLM’s are far from a flawless tool in CS education: they don’t always help, they often hurt, and while they can be a tool for teaching, they are a fraught one.
Session: Inclusion
Marina Krupitskaya (Brooklyn College) started our last session, sharing about how to teach accessible design for cancer survivors with a chemo brain accessibility game. She found that the game led to more awareness and more accessible designs. After, my wonderful students’ Eman Sherif (University of Washington) presented her work on help-seeking visions from students marginalized in CS, aided by Janet Jiang and Saara Uthmaan. Through a 5 hour futures workshop with 18 students from Black and Latine student organization collectively sharing, visioning, and advocating, we found that students sought models that accounted for equity gaps in their lives, that were psychologically safe, that encouraged agency and autonomy, more formative diagnostic assessment than high stakes evaluation, and more identity and culturally-aligned opportunities to build community. The (unexpectedly) last talk was Xinying Hou (University of Michigan) spoke on visioning by high school girls minoritized in CS learning, finding that students felt peer and online support was fast, but unreliable, teachers were more reliable but harder to access, and supplemental tutor support was too mechanical. The essence of the paper’s findings was simply a need for high quality feedback in any form, aligned with basic learning sciences principles.
The bulk of this session was about formative support for learning, particuarly for students often excluded in CS learning. I think I learned from both Xinying’s talk and our own that so many better forms of teaching are possible; it’s likely just a matter of will and resources to build them.
Session: Awards
Immediately after the last talk, we went into the award session, where four papers were recognized:
- Honorable Mention: “Why Put in This much Effort? How AI Availabiliy Shapes Students’ Motivation in Introductory Programming,” Keith Tran, Colton Harper, Thomas Price.
- Honorable Mention: “Not Built for Us: Marginalized Students’ Visions for Help-Seeking in Computing Education,” Belén Edgar, Eman Sherif, Janet Jiang, Saara Uthmaan, Amy J. Ko
- Best Paper: “Both a Powerful Tool and also a Trash Fire”: Investigating the Outlooks, Practices, Challenges, and Debates of Computing Education in the Digital Humanities,” Anna Sollazzo, Quintin Cutts
- Lasting Impact Award: “The Recurring Rainfall Problem,” Kathi Fisler
Congratulations to the authors, especially my Ph.D. students Eman and Janet and my undergraduates Belén and Saara, for pushing the boundaries of the field.
Thanks to all of the organizers for the cozy week in Sweden! And thanks to next year’s organizers, who are taking us to Sacramento, CA for a warm summer week on the best coast next summer.
By the end of the week, I was somehow both rested and wrecked, inspired and defeated, and peaceful and perplexed. The week away from the United States was restorative; the drain of travel and conferencing was not. The opportunity to celebrate a year of research with my collaborators and students was joyful; the dread of returning home that crept closer each day was foreboding. The growth of work at ICER with a liberatory frame was inspiring; the continued obsession with LLMs and lecture optimization was numbing. None of this left me any less inspired: I think public literacy of computing has never been more important, and I’m so excited about the community we’ve grown to help shape it. But heading home I felt the full weight of the work ahead.
Despite all of those mixed feelings, traveling for the week was a joy. Prior to the conference, I had a lovely few days off adventuring with Adrienne, riding bikes, eating quirky food, listening to surprising playlists of 90’s R&B, 60’s Europop, and the occasional jazz standard, immersing myself in dark caverns in museums and gothic parks of salamanders and slime, and absorbing the many peculiar traditions of Swedish past, present, and future. I am a creature of novelty, and so even when I found something in Sweden I do not want to do again, I am forever grateful to have tried it. Thank you Sweden for your hospitality, your shameless pride in oddity, and your beautiful summer days.
farväl och adjö, må vi ses snart igen
