AI and Design Monthly Scholarly Watch (October–November 2026)

Related: AI and Design Monthly Scholarly Watch (September–October 2026) (previous monthly watch) / AI and Design Monthly Scholarly Watch (August–September 2026) (the one before) / AI and Design Monthly Scholarly Watch (July–August 2026) (second watch) / AI and Design Monthly Scholarly Watch (June–July 2026) (first watch) / AI in the Design Industry: An Academic Review (2026) (literature review of AI use across the design field) / AI Adaptation in Design Education — The Current State and Structural Challenges of Curriculum Reform (design education curriculum reform) / Designer Careers That Will Generate Value over the Next Five Years — Academic Review (2026) (designers’ professional value).

Last month’s watch set practice figures (AI use drops off in later workflow phases, and time savings go to product managers) beside education findings, positive and negative, measured on different scales. This window (October 6 to November 2, 2026) covers 12 entries: two experiments and benchmarks on where AI enters the making process, two qualitative studies of collaboration, two on design as an occupation, and six on education. The earlier question of where practitioners use AI is now asked experimentally as a question of when AI should enter.

When AI enters changes team creativity

Huang et al. randomly assigned 148 three-person teams (444 participants) to five schedules of access to the same language model (2026). The task was a 40-minute product-design exercise with a 15-minute search stage, a 15-minute selection stage, and a 10-minute review stage. The two focal conditions differ only in whether the same 15 minutes of access fall in search (early access) or in selection (late access). Teams made 14.9 and 13.8 queries on average, so usage was about the same. Expert-rated creativity was nevertheless 0.356 points higher under early access than under late access (Hedges’ g = 0.62, Welch p = 0.018). Late access produced no detectable gain over working without AI, whereas early access was 0.486 points above the no-AI condition (g = 0.83).

The authors trace the difference through platform logs. Teams with early access grew the number of distinct ideas from about 4 to about 20, roughly fivefold, before the screening gate, then pruned them to fewer than three proposals. Teams with late access entered the selection stage with the same small pool as teams without AI and searched again: 29% of their selection-stage queries asked for new options, against 11% under continuous access. They submitted 73% more proposals than teams without AI, yet creativity ratings did not rise; the authors describe the gain as breadth rather than quality. When language models served as evaluators, they reproduced the effect of having AI at all but missed the cost of its timing.

The participants were students from undergraduates to doctoral candidates with a mean age of 21.6, and the task was a single one. This is a preprint, and no peer-reviewed version was found. Applying it to professional designers’ workflows needs further confirmation. Even so, last month’s survey by Tajarmakan et al. found that adopters used AI in only 2.89 phases on average and withdrew from the later phases. If the withdrawal falls near the selection stage, it would resemble the “late access” of this experiment. That is an inference from placing two studies side by side; the survey does not show which choices respondents were making at each phase.

If the ideas AI adds are alike, increasing their number is worth less. Rusia et al. measured this for UI generation with Design Creativity Bench (2026). Fifteen models received pairs of short briefs with the same goal and different product domains, and the benchmark scored originality (distinctness from other models’ designs for the same brief), creative range (how much a design changes when the domain changes), and appropriateness (the share of the brief’s acceptance criteria met). Originality was 0.592 for model–model pairs (95% CI [0.582, 0.602]) and 0.764 for human–model pairs. Creative range was 0.581 for the models and 0.902 for human designs. Appropriateness exceeded 90% for every model; the best reached 99.2%, against 98.0% for human designs. The authors conclude that default output is appropriate but more repetitive than the human baseline.

All the authors work for the company that develops UI-generation products (Kombai Inc.), and the paper discloses this interest. The live leaderboard is also run by that company. The human baseline consists of designs that several designers adapted from public inspiration libraries to the briefs, not designs conceived from a blank page. It is best read as an unreviewed benchmark value awaiting independent replication.

In collaboration, AI can be a bridge or a wall

Last month’s Zhang and Wang, and Stewart et al., showed AI entering pairings of designers and non-designers. This month two studies look at such pairings more closely.

DuoSketch by Shi et al. lets a pair work on a shared canvas and a live transcript, each member with a separate AI (2026). An exploratory qualitative study of 12 pairs (24 participants) in Singapore found three challenges: AI proposals did not settle into fit with the shared design purpose, shared meanings were not expressed in AI responses, and agreed changes were missing from later outputs. Pairs navigated these through bridging work in two directions: pulling AI output toward the shared design (reworking features, developing uses, assessing fit with requirements) and passing the shared design to the AI (supplying materials, explaining meanings, communicating decisions). It is a work-in-progress paper with a small sample, from the same group as last month’s CommSketch.

Kim et al. combined semi-structured interviews and a participatory design study with 10 narrative writers and 10 visual designers in game development (Proceedings of the ACM on Human-Computer Interaction, CSCW 2026). Generative AI is reported to go beyond individual assistance and mediate collaboration by aligning creative intent, sustaining shared understanding, and easing friction among creators. Participants also worried that it may reduce communication among creators, undermine mutual trust, and reshape interpersonal dynamics and collaborative processes. Where DuoSketch found the bridging work on the human side, this study asks what is lost when AI takes over the bridging role. Neither study compares conditions with and without AI; both treat pairs or differences in expertise as the unit.

Design as an occupation: scaffolding in exchange for standardization

Wadinambiarachchi et al. interviewed 20 professional UX, UI, and graphic designers, most from the Global South, and analyzed the data with reflexive thematic analysis (International Journal of Human-Computer Studies, 2026). The authors report a “paradox of access”: AI gives participants essential scaffolding while imposing a “standardisation tax” that constrains situated cultural expression. The full text was not available, so the per-country breakdown and the concrete episodes behind the standardization tax are unchecked; this summary stays within the abstract. The repetitiveness of default output measured by Design Creativity Bench might be this tax seen from the output side, but the object and method differ, so it cannot be called the same phenomenon.

Huang (2026) argues from 21 interviews with UK film, games, and live-performance practitioners that generative AI is normalizing a mode of labour that demands continual retraining, workflow reconfiguration, and skill upgrading (Media, Culture & Society). The demand is unevenly distributed and its rewards uncertain; AI may widen what practitioners can produce without widening what they can claim in bargaining power, career stability, or recognition. This is not design-specific research, and its full text was not available. It is recorded as peripheral literature.

Education: how self-beliefs move, where students draw the line, and framework proposals

In a cross-sectional survey of 829 art and design undergraduates (40 classes, two institutions in China), Han and Cao found that self-reported prompt literacy was positively associated with learning engagement (β = 0.54), and engagement with both creative self-efficacy (β = 0.62) and creative personal identity (β = 0.47) (Frontiers in Psychology, 2026). The prompt literacy–engagement association was weaker among students with higher cognitive dependence, although all interaction effects were small. The authors themselves note that the data are cross-sectional and self-reported and call for longitudinal and behavioral tests.

The diary study of Li et al. supplies the longitudinal side (Frontiers in Psychology, 2026). They followed 260 Chinese design undergraduates for 14 instructional days, collecting 3,558 person-day observations analyzed with Bayesian dynamic structural equation modeling. Higher-than-usual AI use on one day was followed by higher design-learning self-efficacy the next day (β = 0.043, 95% CrI [0.006, 0.077]), and higher-than-usual self-efficacy by more AI use the next day (β = 0.059, [0.023, 0.095]). The reciprocity is small in both directions. The association from AI use to next-day self-efficacy was stronger among students reporting more baseline hassles (β = 0.458). Last month’s Xiao et al. reported a drop in confidence under the AI condition in a seven-team architecture studio comparison. The diary study shows an association of the opposite sign, but the former is a pre–post comparison and the latter day-to-day within-person variation, so one does not refute the other.

Zhu’s study has two parts (Frontiers in Psychology, 2026). Study 1, a structural equation model of 450 art students, found trust in AI the strongest predictor of behavioral intention (β = 0.425), with the model explaining 59.7% of its variance. Study 2 randomly assigned 80 students, 40 per group, to receive metacognitive prompts or not. Expert ratings favored the prompted group on semantic content (d = 0.67), visual style (d = 1.30), and spatial composition (d = 0.63), and its outputs were also more diverse. Study 2 does not test the trust-mediation pathway (the author states this), and the participants are art students; transfer to design education needs separate confirmation.

John Luan and Murphy asked 48 Digital Media students (of a population of 120) at an Australian university which forms of AI involvement they found acceptable (Education Sciences, 2026). Respondents were concerned about AI training practices (93.6%), reduced employment opportunities (89.4%), the ethical acceptability of AI-supported coursework (83.0%), environmental effects (80.9%), and possible reductions in creative-skill development (76.6%); the author states that these percentages represent perceived risks, not demonstrated outcomes. Students separated bounded technical, corrective, and organisational assistance from involvement that would displace ideation, research, planning, understanding, or capability building, and the author proposes “conditional legitimacy” as a provisional framework for that distinction. The sample is small and the full text was not available, so this summary stays within the abstract.

Fatnassi (2026) proposes a framework for the product design studio, “regulated co-intelligence”, consisting of four guiding principles, a five-phase studio model, and mechanisms regulating AI use (The International Journal of Design Education). The abstract positions it as a conceptual basis for future empirical research. Kayabaşı (2026) compared 100 articles, 20 each from graphic design, visual communication design, animation, game design, and UI/UX, following PRISMA 2020 (ODÜSOBİAD). The role of AI differs by field; in UI/UX, for example, it contributes to modeling user behavior and to prototyping. Originality, copyright, and the transformation of designer identity recur as challenges in every field. The full text of neither was available.

Gaps (open issues)

First, the “when to bring AI in” experiment is limited to a student sample and one task. No study in this window recorded workflow phase and AI arrival time in professional practice to check whether Huang et al.’s result holds there.

Second, the only figure on the repetitiveness of default output is an unreviewed benchmark by authors with a stake in the outcome. Independent evaluation and replications that construct the human baseline differently are needed.

Third, the six education papers rely mainly on self-report surveys or small samples, and none follows the quality of design-course deliverables longitudinally. The diary study of Li et al. tracks self-efficacy by day and Zhu’s experiment rates deliverables, but the former does not look at deliverables and the latter does not look at continuing processes.

Fourth, relevant DRS 2026 papers and the ACM Digital Library list for C&C ‘26, carried over from the previous watch, were again not searched.

Reading notes

No major conference fell inside this window, so entries were collected from arXiv (including cs.HC) and from journal articles in Crossref with publication dates from 2026-10-06 to 2026-11-02. Of the 12 entries, three are arXiv preprints (Huang et al., Shi et al., Rusia et al.), none peer reviewed. Kim et al. has both an arXiv version and the peer-reviewed ACM version (CSCW 2026), and the bibliography follows the latter. Full texts were obtained and checked for the three Frontiers papers (Han and Cao, Li et al., Zhu) and the four arXiv papers. For Wadinambiarachchi et al., Huang (2026), John Luan and Murphy, Fatnassi, and Kayabaşı, full texts were unavailable and the summaries stay within the abstracts. Source tracking is recorded in the Provenance and verification-evidence sections of the corpus.

References

All accessed on November 2, 2026.

• Huang, D., Yu, S., Chen, F., Zhou, W. “Too Late to Prune: Generative AI, the Screen, and Team Creative Decisions.” arXiv (not peer reviewed). https://arxiv.org/abs/2610.14984
• Rusia, A., Bhole, A., Gupta, P., Dey, D. “Design Creativity Bench: Measuring creativity in LLM-Generated UI.” arXiv (not peer reviewed, v2). https://arxiv.org/abs/2610.11539
• Shi, W., Lim, D., Quek, G., Choo, K. T. W. “DuoSketch: How Pairs Navigate Challenges in AI-Supported Collaborative Design Ideation.” arXiv (not peer reviewed). https://arxiv.org/abs/2610.10249
• Kim, H., Kim, J., Choi, Y. J. “Generative AI as a Mediator in Creator Collaboration: Challenges, Design Opportunities, and Concerns.” Proceedings of the ACM on Human-Computer Interaction, 10(6), CSCW092. https://doi.org/10.1145/3816940
• Wadinambiarachchi, S., Waycott, J., Wadley, G. “Asymmetric encounters with AI: Professional designers’ perceptions and integration of AI tools.” International Journal of Human-Computer Studies, 216, 103891. https://doi.org/10.1016/j.ijhcs.2026.103891
• Huang, J. “Self-programmable creative labour: Generative AI and the reorganisation of work in the UK creative industries.” Media, Culture & Society. https://doi.org/10.1177/01634437261494891
• Han, L., Cao, Z. “Learning engagement and creative self-beliefs in generative AI-supported design education: self-reported prompt literacy and dimensions of AI dependence.” Frontiers in Psychology, 17, 1957097. https://doi.org/10.3389/fpsyg.2026.1957097
• Li, Y., Yang, Y., Bao, S. “How generative AI becomes embedded in design learning: a 14-instructional-day intensive longitudinal study of reciprocal dynamics between AI use and self-efficacy.” Frontiers in Psychology, 17, 1975841. https://doi.org/10.3389/fpsyg.2026.1975841
• Zhu, R. “Supporting creative agency: metacognitive control and trust in generative AI art education.” Frontiers in Psychology, 17, 1997631. https://doi.org/10.3389/fpsyg.2026.1997631
• John Luan, R., Murphy, B. “Beyond Acceptance and Resistance: Digital Media Students’ Judgements of Generative AI Assistance and Perceived Creative Replacement.” Education Sciences, 16(10), 1725. https://doi.org/10.3390/educsci16101725
• Fatnassi, M. “AI-Augmented Studio Pedagogy.” The International Journal of Design Education. https://doi.org/10.18848/2325-128x/cgp/a356
• Kayabaşı, Y. “Generative AI in visual design processes: A systematic review and comparative analysis from the perspective of design disciplines.” ODÜ Sosyal Bilimler Araştırmaları Dergisi (ODÜSOBİAD), 16 (ISRIS 2026). https://doi.org/10.48146/odusobiad.1959175