AI and Design Monthly Scholarly Watch (August–September 2026)

Centering on August 4 through September 7, 2026, and adding as an extended window 3 papers from a newly discovered journal issue (published July 29, 2026), 7 peer-reviewed papers and preprints at the intersection of “AI and design” were collected. The scope covers HCI (CHI, DIS, Creativity & Cognition), Design Studies-adjacent venues, design education, and the learning sciences. Ledger details (predatory-journal checks, quality flags, limits of search coverage) are in the corpus (source/review/ai-design-scholar-watch-2026-09/papers.md).

Related: ai-design-scholar-watch-2026-08 (previous monthly watch) / ai-design-scholar-watch-2026-07 (watch before that) / ai-in-design-literature (literature review of AI use across the design field) / design-education-ai-adaptation (design education curriculum reform) / designer-career-value-literature (designers’ professional value).

The previous watch counted 25 papers because its window included ACM Creativity & Cognition 2026, but no conference of comparable scale falls within this window. The count of 7 does not carry forward the momentum of the previous watches on a simple count basis. That thinness, however, reflects the nature of the window rather than a shortfall in search effort. Indeed, the 7 papers collected add new material to each of the two conflicts carried over from earlier watches.

Preserving Controllability and Agency in AI-Assisted Tools

The previous corpus left standing a conflict between a body of work premised on the improvability of AI-assisted tools and a body of empirical studies reporting negative effects of AI mediation. The 3 papers here share, within the tool-research side of that conflict, a single question: how not to let go of control. Zhou et al.’s LegoUI (2026) presents a framework that, rather than having generative AI produce a UI design in one shot, stages the process into UI-DSL blocks carrying provenance information. A technical evaluation on 40 real design prompts reports requirement-extraction accuracy above 95%, and a user study reports improvements in transparency, controllability, and intent alignment over existing one-shot generation tools. Staging the process and making provenance explicit is an attempt to turn black-box generation into a process humans can trace.

Akhtar’s workshop paper (2026) takes up the same problem outside the tool itself, in the setting of qualitative UX data analysis. A comparative analysis using 20 user responses points to the risk that AI-assisted sensemaking flattens the rich patterns in data, and proposes relocating AI’s value from “eliminating subjectivity” to “enhancing intentionality, reflexivity, and explainability.” Where LegoUI designs controllability on the output side of generation, Akhtar identifies a pathway by which designer agency is lost on the input side of interpretation — the same concern illuminated from both ends of the generative process.

Alzouby’s position paper (accepted at a CHIWORK 2026 workshop; [needs primary-source verification]: a timing inconsistency in the submission record remains unresolved), framing both of the above, argues that generative AI should be positioned not as a replacement for human-centered design but as a “partner to think with,” naming deskilling and an accountability vacuum as risks. It stops at a programmatic claim without accompanying design or empirical work, but it names, as a challenge for HCI as a whole, the very problem that both LegoUI and Akhtar are trying to address.

Empirical Evidence from Design Education: What Does AI Reorganize?

The 4 education-cluster papers add design-specific verification to the “dual-mechanism model” (an amplifier under structured instruction, a substitute without it) flagged by the previous corpus. Xu et al.’s correlational study (2026) surveyed 150 students in Chinese vocational design education on the relationship between perceived GenAI-supported learning and creative design performance. The result is not simple. What correlated most strongly with expert-rated creative design performance was learning engagement, while the direct effect of perceived AI support itself on design performance was small. That the depth of engagement with learning, rather than the strength of the felt sense of using AI, governs design performance works to corroborate, in the design-education context, the “structure of instruction” condition set by the previous watch’s dual-mechanism model. The authors themselves, however, note the limits on causal inference from simultaneous measurement and the absence of a comparison condition; this data cannot distinguish whether engagement with AI raises learning engagement or the reverse.

Oral’s qualitative comparative study (DTEIJ, 2026) gives this question of “what AI is effective on” a more concrete mechanism. From observation in an architectural design studio, it argues that generative AI does not simply make production more efficient but reorganizes the decision-making process itself, showing that in AI-mediated making, students lean toward choosing among pre-generated options, and that traceability of decisions and a sense of ownership over the design both decline. This corroborates, at the level of individual students’ decision-making processes, the “compression of collective diversity” and “visual homogenization” reported by the previous corpus, and shows that AI’s negative effect is not mere degradation of output but a transformation of the act of choosing itself.

Yan’s integrative review (DTEIJ, 2026) gives a framework for addressing this transformation. Synthesizing 187 studies from 2022 through 2026, it identifies 5 clusters of GenAI-affected design challenges and 4 pedagogical risks, and proposes “systems stewardship” — an educational goal that shifts the center of gravity of design judgment from individual choice to governance and system-level coordination — as the response. Where Kang’s “generative refusal,” introduced in the previous corpus, was the design of a single tool, Yan’s proposal is a reorganization of the curriculum as a whole, presenting the same prescription at a larger grain. Fatnassi’s curriculum comparison survey (DTEIJ, 2026), covering 7 international institutions, shows that while the degree of integration of AI-related competencies varies widely across institutions, foundational product-design skills remain stable — corroborating that the framework Yan proposes is, in practice, still at an uneven stage of institution-by-institution implementation.

Gaps (Unmet Points of Inquiry)

This corpus, too, carries forward some gaps and opens new ones.

First, the central conflict carried over from the previous watch is still not directly tested. Oral’s qualitative study shows a concrete mechanism — declining traceability of decisions and ownership — but this is a qualitative observation in a single studio, not an experiment that controls for the “branching by presence or absence of structured instruction” the dual-mechanism model predicts. The controllability-by-design improvements the tool-research group (LegoUI and others) presupposes and the accumulating negative-effect reports from design-education practice continue to run as separate bodies of research.

Second, the 3 papers in this watch’s extended window are concentrated in a single issue of a single journal (Design and Technology Education: An International Journal, Vol. 31, No. 2). The possibility that this issue was an AI special issue cannot be ruled out, so it cannot be generalized as a population-level trend.

Third, the search stage confirmed the existence of related papers at DRS 2026 (held June 8–12, 2026) not captured in this corpus — appearing from their titles to address critical themes such as “Small AI,” “archipelagic AI,” and “the hauntology of generative AI” — but there was not enough time to confirm their publication and availability dates, so they were set aside this round. They remain a candidate for inclusion as an extended-window addition in the next watch.

Notes on Reading

This window (August 4 through September 7, 2026) includes no conference of the scale of ACM Creativity & Cognition or DRS, which is the principal reason for the drop in count from the previous watch (25 papers). The ACM Digital Library’s C&C ‘26 proceedings listing again returned a 403 and could not be confirmed, continuing the pattern from the previous two watches. Design and Technology Education: An International Journal, adopted here as an extended-window addition, is a venue discovered for the first time in this watch; it is an open journal published by Liverpool John Moores University (ISSN 1360-1431/2040-8633) whose peer-review process and ethics policy were confirmed, but the judgment that it is not a predatory journal rests on a single check made this round. All 3 arXiv preprints (SEP26-01, SEP26-02, SEP26-04) are not peer reviewed, and one of them (Alzouby) has an unresolved inconsistency between its arXiv ID and its submission-date metadata. Details of provenance tracking are recorded in the Provenance section of the corpus.

References

All accessed September 7, 2026.

• Zhou, Yuan, Xu, Chen, Wen, Pan, Xu, Zhang, Quigley, Xing, Mohammadi. “LegoUI: Designing with UI-DSL Bricks to Balance Transparency and Controllability.” arXiv (not peer reviewed). https://arxiv.org/abs/2608.04293
• Akhtar, M.H. “Between Algorithm (AI) and Intuition (Human): Preserving Designer Agency in AI-Assisted Sensemaking of Qualitative UX Data.” arXiv, accepted at the CHI 2026 “Sensemaking and AI” workshop (not peer reviewed). https://arxiv.org/abs/2608.28420
• Xu, X., Wang, C., Fu, S., Tong, J., Jin, D. “Psychological associations between perceived generative AI-supported learning and creative design performance in vocational design education: a within-course correlational study.” Frontiers in Psychology, 17. https://doi.org/10.3389/fpsyg.2026.1962146
• Alzouby, I. “From Human-Centered Design to Human-AI Collaboration: Why the Future of HCI Still Starts With People.” arXiv, position paper accepted at a CHIWORK 2026 workshop (not peer reviewed). https://arxiv.org/abs/2608.07482
• Fatnassi, M. “Reconfiguring Product Designer Competencies in the Age of Artificial Intelligence: Towards a Pedagogy of Mediation in Project Workshops.” Design and Technology Education: An International Journal, 31(2), 98-126. https://doi.org/10.24377/DTEIJ.article3479
• Oral, M. “Spatial and Conceptual Intuition in Human–AI Interaction: A Qualitative Comparative Study in the Architectural Design Studio.” Design and Technology Education: An International Journal, 31(2), 146-177. https://doi.org/10.24377/DTEIJ.article3531
• Yan, H. “From Output Production to System Stewardship: A Curriculum Framework for AI-Mediated Design Capability in Design and Technology Education.” Design and Technology Education: An International Journal, 31(2), 178-195. https://doi.org/10.24377/DTEIJ.article3542

Unverified Items

• SEP26-01 (LegoUI): not peer reviewed preprint.
• SEP26-02 (Akhtar): not peer reviewed workshop paper.
• SEP26-04 (Alzouby): not peer reviewed position paper. The arXiv ID 2608.xxxxx and the submission history’s submission date (2026-06-09) are inconsistent; the possibility remains that the actual submission falls outside the central window.
• The 3 extended-window papers (DTEIJ26-01 through 03): all 3 come from a single issue of a newly discovered journal, and the possibility that it was an AI special issue cannot be ruled out.