Designing in Uncertainty: GenAI and Creative Agency in a Graduate Seminar

Synthesis: Adeloju, Mishra, McCaleb and Oster (2026) report a 14-week graduate seminar, Education By Design, that treated GenAI not as a tool to be regulated but as a creative partner in design work, organized through the Five Spaces of Design in Education framework — Artifacts, Processes, Experiences, Systems and Culture. The course's distinctive move was compositional: 25 students including high schoolers, master's and doctoral students and university faculty worked as equals in one public space, with no restrictions on GenAI use provided it was disclosed and cited. Assignments were layered at micro (weekly reflections, a recurring "Good and Bad Designs" observation task, eight student-made session documentaries), meso (six satirical AI commercials, twelve theory explainer reels, a Vice Provost's distributed-learning challenge that produced a 23-page proposal) and macro scale (a semester-long individual vibe-coded project presented at a student-organized public "AI Carnival"). This is a practitioner report, not a study: the evidence is selected student reflection quotes and instructor documentation, with no comparison group, no outcome measures and no coded dataset, so it supports a design model rather than a claim about learning gains.

Key Findings

1. GenAI was framed as a designed artifact, not a neutral tool. Students analyzed AI-generated experiences alongside physical designs in the "Good and Bad Designs" task, learning to see Generative AI itself as carrying embedded assumptions they could critique.
2. The course removed the permission question and kept the disclosure obligation. GenAI use was unrestricted and actively encouraged across all assignments, conditional on full disclosure and proper citation — the reverse of the default institutional posture the authors describe as moving between prohibition and uncritical adoption.
3. Deliberate heterogeneity flattened hierarchy. High school students, master's and doctoral students, and faculty worked in the same room as equals, which students described as the main source of comfort in taking creative risks.
4. Assignments escalated across three scales. Micro tasks built design literacy and observation habits, meso group projects introduced authentic external challenges, and macro individual projects demanded sustained iteration over a full semester.
5. Authentic external stakes produced the reported shift. The Vice Provost's distributed-learning challenge — redesigning access for populations historically excluded from higher education — was named by a student as "the turning point" from insightful to transformative.
6. Public presentation was part of the design, not an assessment afterthought. The course closed with a student-organized AI Carnival for high schoolers, teachers, faculty, researchers and community members, which students described as "systems design in real time."
7. The framework was used descriptively, not analytically. The authors state that the Five Spaces guided documentation of what the course produced rather than serving as a coding scheme, and that the boundaries between spaces are descriptive rather than definitive.

Designing for uncertainty rather than for a stable tool

The paper's argument is that both dominant institutional responses are category errors. Prohibition assumes GenAI can be kept outside the learning environment; uncritical adoption assumes it can be dropped in. Neither engages with the fact that GenAI was not designed for education, so making it educational is itself an act of design. That reframing changes what a course is for: if the technology shifts faster than any policy can track, the durable outcome is not a correct position on GenAI but a disposition for working without one. The authors ground this in Buchanan's and Rittel and Webber's account of wicked problems — educational problems that resist stable definition — and in the design tradition's claim that such problems are addressed through iterative, one-shot interventions rather than through analysis followed by implementation.

The Five Spaces framework gives that argument a structure. The Artifact space holds the materials that mediate learning; Process holds sequencing and facilitation; Experience holds how learners encounter and make meaning; Systems holds institutional arrangements and policy; Culture holds shared values and assumptions. GenAI, on this account, is an artifact that simultaneously reshapes process, experience, system and culture — which is why a course that treats it as a single tool misses most of what it changes. The assignment sequence enacts the framework: micro-level observation tasks train attention on artifacts, meso-level external challenges force engagement with systems, and the heterogeneous composition deliberately works on culture. Design Thinking here is not a five-step procedure but a way of encountering a problem whose solution is not yet knowable.

What the layered assignments actually required

The micro level is the most transferable piece. Each week students submitted questions and curiosities drawn from the readings before class, which then seeded discussion — a reading-responsiveness structure rather than a comprehension check. A rotating group presented personally observed examples of good and bad design, with photographs required and online sources disallowed, which the authors describe as perceptual training: several students reported that the activity changed how they moved through ordinary environments. A second rotating group served as class documentarians, producing eight creative session documentaries across the term as podcasts, poems, videos, websites, infographics and mind maps, made with GenAI. The design choice worth borrowing is that GenAI was the medium of documentation, not a topic of discussion — students were continuously making something with it rather than talking about it.

The meso level added external audiences and real constraints. Groups produced six 45-second commercials for fictional AI companies, deliberately in satirical, comedic or dystopian registers, which required students to use GenAI to make an artifact and simultaneously interrogate the artifact's assumptions. Twelve short reels explained design theorists for a general audience, translating complex positions into short narrative. The distributed-learning challenge then arrived from the Vice Provost's office with a two-week deadline and a 23-page proposal as its deliverable. The macro level carried the individual work: each student iteratively prototyped an AI-assisted artifact for a self-identified educational problem, ending with a minimum viable product and a reflective paper. The authors' point about "minimum viable" is deliberate — project work in design does not converge on a final solution, and the reflection paper asks students to account for the iterations rather than defend an outcome.

What this means for practice

• Instructors. Replace the permission question with a disclosure and citation requirement, then design assignments in which GenAI is the medium of production rather than the object of discussion.
• Instructional designers. Sequence work across at least two audience scales: internal peer work first, then a genuine external deliverable with a real stakeholder and a real deadline.
• Curriculum designers. Build heterogeneity in deliberately — mixed levels and roles in one room is a design decision that changes what students can learn from each other, not an enrollment accident to be managed.
• Faculty developers. Use the micro-level observation task (personally observed good and bad designs, no online sourcing) as a low-cost entry point; it requires no technology and trains the critical stance the rest of the course depends on.

Limitations

• This is a practitioner report, not research: the authors describe student reflections as "practitioner evidence" selected to illustrate representative work, with no coding scheme, no outcome measures and no comparison condition.
• The sample is 25 students in one 14-week seminar at one institution in one semester (Fall 2025), and the course was team-taught by the authors, who also selected the quotations.
• The Five Spaces framework organizes the reporting but was not used as an analytical coding scheme, so the mapping from evidence to space is the authors' interpretive judgment.
• Student quotations are drawn from required end-of-semester reflection papers, which are graded course artifacts and therefore not neutral testimony.
• No claim about learning outcomes can be drawn: the report documents activities, produced artifacts and student accounts, not measured change.

Citation

Adeloju, E., Mishra, P., McCaleb, L., & Oster, N. (2026). Designing in Uncertainty: GenAI and Creative Agency in a Graduate Seminar. Journal of Instructional Design and Technology, 1(2), 1–7.