Scope and Method of Derivation

This note is not a new literature collection but a synthesis of existing notes. The evidence base: the four lineages of failure-embedded learning design (Designing Failure Into Learning), the debate with the opposing school and its boundary conditions (the debate note), the harms of delegated thought generation by generative AI (the offloading note), the robustness assessment of technology-education harms (the harm-patterns note), and scenario-based learning with GBS (the scenario note, the GBS literature map). The state of curriculum reform and skill identification is covered by design-education-ai-adaptation, and AI-premised competency measurement by ai-augmented-competency-measurement; this note addresses the learning-design principles that sit upstream of both. A draft underwent one critical review by an academic critic on the design-education canon (educational and learning-sciences theory), and its findings (counter-evidence to the central distinction, limits of an analogy, missing perspectives) are incorporated; canon citations were not verified at page level and carry citation-needed flags.

Terms first. Craft education here means education whose learning objectives are production skills (drawing, rendering, mockup building, implementation, tool mastery). Design education means education whose learning objectives are problem framing, the exploration and generation of alternatives, and evaluation and judgment (the ability to judge what is good design and articulate why). This distinction, however, is a working assumption, not a settled premise. Whether craft and design judgment can be severed at the level of learning objectives is itself an issue challenged by the educational canon (Issue 6), and Issues 1, 3, and 5 below are conditional arguments standing on this assumption.

Issue 1: Transferred to Design, the Offloading Criterion Strikes the Vital Point

The design principle derived from the harm-patterns note was: harm appears when the delegated processing is itself the current learning objective. Calculators delegated instrumental computation and caused no harm (except grade 4, mid-acquisition); answer-providing generative AI delegated novices’ thought generation, the learning objective, and caused harm.

Transferring this criterion to design education exposes a structural asymmetry. Under the working assumption, AI delegation of craft (rendering, implementation, mockups) is calculator-like, freeing resources by delegating processing outside the learning objectives. But what generative AI delegates by default in a design context is precisely the generation and exploration of alternatives — design education’s learning objective itself. In programming education AI delegated implementation and design could remain; in design, the core function of generative AI (prompt in, alternatives out) delegates exploration directly. By the offloading criterion, design education is the educational domain where generative AI’s default behavior strikes the learning objective directly. The finding recorded in design-education-ai-adaptation (reflective engagement drops when AI substitutes for early ideation) reads as an observation of this structure. Still, this diagnosis is a testable hypothesis (“design education takes the hit harder than other fields”), extrapolated without validation in ill-structured domains (Issue 2).

Issue 2: The Limits of Extrapolation. How to Define Retention and Transfer for Design Problems

Both the harm evidence (+48% practice, −17% exam) and the failure-design evidence come from mathematics, physics, programming, and essay writing — domains with definable answers or rubrics. As noted in the offloading note, no validation exists in domains without unique answers. Design problems are placed among the most ill-structured in the typology of problems (Jonassen 2000), and the extrapolation faces a double difficulty.

First, measuring “what remains when the AI is removed” requires an operational definition of what should remain; exam scores suffice for mathematics, but measuring “the ability to judge well” in design is unestablished (connecting to ai-augmented-competency-measurement). Second, the boundary conditions of failure design (generating contrasting cases; instruction scaffolded on learners’ solutions) were formulated for domains with canonical answers, and what “failure” and “consolidation” mean without them is not obvious. Moreover, the very framing “what remains in the individual without AI” presumes learning as individual acquisition; from the stance that treats judgment as participation in a community of practice (Lave & Wenger 1991, citation unverified), the framing itself is contestable. Design education stands not on the borrowing side of this evidence but on the side that should supply the validation for ill-structured domains.

Issue 3: The Studio Already Institutionalizes Post-Failure Engagement. AI Undermines Its Precondition

The largest gap Designing Failure Into Learning found in industry services was post-failure consolidation. Design education is exceptional here: the studio crit is an institution in which, after the student’s attempt (the work), an expert engages on the basis of that attempt — a two-phase structure (attempt first, expert engagement second) institutionalized for decades independently of productive failure theory.

This correspondence must be marked as a partial analogy, however. PF’s consolidation phase presupposes convergence to canonical solutions, and design has none (Issue 2). Schön’s reflective practicum describes the coach-student exchange not as consolidation toward a correct answer but as something closer to reciprocal exploration (citation unverified). The crit shares with PF only one point: the learner’s own attempt serves as the scaffold for expert engagement. No empirical study analyzing the crit within the PF framework was found — an absence that can also be read as a sign the correspondence may not hold.

The diagnosis that early AI use strikes the studio’s vital point nonetheless stands. The crit’s precondition is that its object is the trace of the student’s own exploration. A crit of AI-generated alternatives becomes connoisseurship of AI output rather than scaffolding of the student’s thinking: the engagement phase (crit) survives while the attempt phase it should engage (the student’s generative struggle) disappears. And this is not solely cognitive: that the critiqued work is one’s own is also the basis of authorial identity and motivation (ownership), which the connoisseurship of AI output erodes at the same time (Issue 7).

Issue 4: The Assessment Dissociation. Portfolios Measure Performance, Not Learning

The learning-performance dissociation (Contention 3 of the debate note) strikes design education’s assessment institution directly. The traditional assessment is the portfolio — artifact quality. Under AI co-use, artifact quality is “performance with the AI present” and can dissociate from capability: the same intervention that raised practice scores 48% lowered exam scores 17%.

The issue is therefore whether assessment can shift from the artifact to the designer. Three directions exist: process assessment (traces of exploration, articulated reasons for judgments), a parallel no-AI assessment condition, and frameworks measuring collaboration with AI itself as a competency (agency allocation and output evaluation in ai-augmented-competency-measurement). Process assessment carries an inversion risk flagged in the critical review, though: grading the traces of exploration can turn the space of exploration into one under evaluative surveillance, closing the very space where failing is safe (Freire’s critique of the banking model — who deposits what as the assessed answer; citation unverified). Restricting the collection of process traces to formative feedback, severed from summative judgment, is the condition for avoiding that inversion.

Issue 5: Staged Design Along Two Axes: When, and What

The most convergent point of the debate was that prior knowledge changes the optimal instructional form (expertise reversal), and the AI-harm evidence concentrated on novices and strugglers (illusion of competence, widening gap). Vygotsky’s zone of proximal development (calibrating support to the learner’s developmental level) offers a second theoretical pillar (citation unverified).

Transferred to design education, the curriculum becomes a staged function rather than a ban-or-allow binary — but, as the critical review argued, the time axis (when to allow) is insufficient on its own. An independent axis is needed: what is allowed to be offloaded. “Restricting generative delegation” can mean two different things: (a) not letting the AI make the artifact (keeping making with the student), or (b) restricting AI use as a tool altogether. By Papert’s constructionism (learning is strongest when learners build and manipulate public, manipulable artifacts; citation unverified), (a) is consistent while (b) risks depriving novices of building-to-think. Staged design means specifying, at each stage, what students must keep making — not opening and closing a valve on a timeline.

An unresolved tension remains. The failure-design prescription (protect novices’ generative struggle) and the cognitive-load prescription (novices need scaffolds and direct instruction first; Kirschner et al. 2006) collide most sharply in ill-structured design tasks; protecting struggle cannot be adopted as an unconditional good. And with no study operationalizing “early” (the offloading note’s gap), the coarse basic/advanced split still needs refining into a map of which design judgments are transformative thresholds (threshold concepts; citation unverified).

Issue 6: Can Craft and Design Be Severed? (Examining the Working Assumption Itself)

The working assumption (craft = instrumental processing; design judgment = learning objective) faces frontal objections from the educational canon. The critical review’s mobilized positions organize into opposing predictions.

Against severance. In Dewey’s theory of experience, reflective thought arises from difficulty within action and does not preexist as pure judgment severed from making (citation unverified). In Lave & Wenger’s legitimate peripheral participation, novices enter a community’s standards of judgment through peripheral craft practice, so placing skill outside the learning objectives externalizes the very entry path to judgment (citation unverified). Schön’s back-talk (the material’s response trains judgment) and Papert’s constructionism also ground the making-forms-judgment side (citations unverified).

Potentially for severance. One reading of Vygotsky’s tool mediation holds that a tool’s function can be internalized without manual mastery; if so, design judgment might be internalized through AI as a mediating tool without craft skill (citation unverified).

The issue thus lacks not only evidence but features opposing theoretical predictions, which makes it observable: control the quantity of making and track the development of design judgment (if the anti-severance side is right, students with little craft experience should lag in judgment development). This note places no conclusion, only the practical implication: there is no ground for declaring craft offloading harmless (despite the criterion’s in-principle application), and identifying and retaining the minimal making that feeds judgment formation, at least in the basic stage, hedges against both predictions.

Issue 7: Beyond Cognitive Reduction: Ownership, Community, Equity

The critical review identified perspectives missing from the draft; they stand as an issue of their own. The preceding issues discussed AI’s impact in cognitive vocabulary (offloading, retention, transfer), but design learning is not driven by cognition alone.

First, ownership and identity. Studio learning involves forming an authorial identity, and whose work it is regulates motivation; the idling of crits on AI-generated work (Issue 3) is an ownership problem as much as a cognitive one, and affect was the thinnest-evidenced contention on both sides of the failure-design debate.

Second, community and power. The crit is at once a scaffolding institution and a site of selection and socialization where instructors define whose judgment is legitimate; idealizing it as a consolidation phase overlooks its excluding function, and the surveillance inversion of process assessment (Issue 4) belongs to the same family.

Third, equity. AI’s harms are not distributed evenly: transferring the “widening gap” observation (high performers accelerate, weak performers depend and mask) to design education, staged design (Issue 5) carries the question of who gets to climb the stages. The benefit of lowered exploration barriers and the harm of dependence-driven divergence are two faces of the same tool, and whether staged design reproduces existing hierarchies deserves independent monitoring.

Contention Table (Unresolved Disputes for Design Education)

Merged with the critical review’s table. Cited, unresolved conflicts over consensus.

Axis	Position A (source)	Position B (source)	Evidence needed
Can craft and design judgment be severed?	Severable by learning objectives (this note’s working assumption; calculator analogy)	Experience and participation constitute skill and judgment inseparably (Dewey; Lave & Wenger; citations unverified)	Longitudinal studies controlling quantity of making and tracking judgment development
Should novices’ generative struggle be protected?	Yes (failure-design lineage)	Novices need scaffolds and direct instruction first (Kirschner et al. 2006)	Pre-registered RCTs of struggle-first vs scaffold-first on ill-structured design tasks
Does judgment develop without making?	Making forms judgment (Papert; Schön; citations unverified)	Internalizable via tool mediation (one reading of Vygotsky; citation unverified)	Empirical work on craft-experience quantity and evaluative-judgment development
Is the crit PF’s consolidation phase?	An institution sharing the two-phase structure (Issue 3’s limited correspondence)	Reciprocal exploration without canonical solutions, not consolidation (Schön; citation unverified)	Empirical analysis of crits in ill-structured domains within the PF framework
Does assessing exploration traces open learning?	Shift to process assessment (Issue 4)	Grading traces closes the exploration space under surveillance (Freire; citation unverified)	Observation of exploration behavior and self-disclosure before/after process assessment
Can co-exploration with AI be a learning objective?	Restrict as delegation of exploration in the basic stage (Issue 1)	Learning by co-doing is itself the new objective (design-education-ai-adaptation)	RCTs measuring no-AI transfer on ill-structured tasks

A Warning Against Over-Adaptation, and a Conclusion Not to Adopt

One lesson carries over from the harm-patterns note: many famous educational harms collapsed under peer-reviewed refutation, and over-adapting to harm panics costs the benefits. A wholesale AI ban in design education would likely fit that pattern.

The opposite simplification is equally unadoptable. The critical review named the conclusion not to adopt: “restrict generative AI in the basic stage and protect generative struggle, and design education is safe.” That ignores the cognitive-load prescription (novices need scaffolds most) by treating struggle as an unconditional good, and risks stripping building-to-think in the name of restriction. This note’s issues justify neither shutting AI out nor prescribing struggle, but designing what is offloaded, when, for whom, and what students keep making — with the conflicting evidence in view. The generative-AI harm originals have not yet faced replication; these issues should be rewritten as that evidence updates.

Gaps

• Validation of AI harms and failure design in ill-structured design problems (Issue 2): zero empirical work; Issues 1, 3, 5 rest on extrapolation.
• Direct tests of whether design judgment develops without craft experience (Issue 6): the canon supplies opposing, testable predictions.
• Empirical studies analyzing the studio crit within the failure-design framework — the same root as the GBS literature map’s gap (no explicit GBS application to design studios).
• An operational definition of “early” for design, and refinement of stages via threshold concepts (Issue 5).
• Measurement of the dissociation between portfolio assessment and capability retention (Issue 4).
• Empirical work on ownership, motivation, and equity in AI-co-used design learning (Issue 7).

Unverified Items

• As a synthesis note, unverified items of the cited empirical work are inherited from the referenced notes’ corpora.
• Educational-canon citations derive from the academic critic’s memory-based references, unverified at page level: Dewey (1938) continuity of experience; Lave & Wenger (1991) inseparability passages; Papert (1980) objects-to-think-with; Vygotsky (1978) tool mediation (whether “internalization without manual skill” is supported by the original is especially open); Schön (1983/1987) back-talk and reciprocal exploration; Meyer & Land (2003) threshold concepts; Freire (1970) banking-model critique.

---

References

All empirical grounding rests on the following notes and their corpora (full bibliographies at each note’s end).

• Designing Failure Into Learning — the four failure-design lineages and industry implementations
• The Debate Over Designing Failure Into Learning — the CLT debate and boundary conditions
• Does Early Use of Generative AI Inhibit the Formation of Thought? — harms of delegated thought generation
• Patterns of Technological Harm to Education — robustness assessment of claimed harms
• The GBS literature map — GBS’s seven components and design contexts
• Scenario-Based System Design Learning — industry scenario learning against GBS
• design-education-ai-adaptation — the state of design education’s AI adaptation
• ai-augmented-competency-measurement — AI-premised competency measurement

Individual works cited directly (educational canon via the critical review; many unverified at page level):

• Jonassen, D. H. (2000). Toward a Design Theory of Problem Solving. Educational Technology Research and Development, 48(4), 63–85. https://doi.org/10.1007/BF02300500
• Kirschner, P. A., Sweller, J., & Clark, R. E. (2006). Why Minimal Guidance During Instruction Does Not Work. Educational Psychologist, 41(2), 75–86. https://doi.org/10.1207/s15326985ep4102_1
• Dewey, J. (1938). Experience and Education. Kappa Delta Pi. (citation unverified)
• Lave, J., & Wenger, E. (1991). Situated Learning: Legitimate Peripheral Participation. Cambridge University Press. ISBN 9780521423748.
• Papert, S. (1980). Mindstorms: Children, Computers, and Powerful Ideas. Basic Books. (citation unverified)
• Vygotsky, L. S. (1978). Mind in Society. Harvard University Press. (citation unverified)
• Schön, D. A. (1983). The Reflective Practitioner. Basic Books. ISBN 9780465068784. / Schön, D. A. (1987). Educating the Reflective Practitioner. Jossey-Bass. (citation unverified)
• Meyer, J. H. F., & Land, R. (2003). Threshold Concepts and Troublesome Knowledge. (citation unverified)
• Freire, P. (1970). Pedagogy of the Oppressed. Herder and Herder. (citation unverified)