The cycle that began with [YOU] on AI confronts the problem of constructive friction across every domain it touches. The builder who asks an AI for code and receives it has gained an artifact without gaining the understanding that writing the code would have produced. The student who asks an AI for an explanation receives an answer without undergoing the inquiry that the search for the answer would have demanded. The organization that routes its analysis through AI systems accumulates outputs without developing the institutional judgment that producing those analyses would have exercised. In each case, the friction—once the price of understanding—has been removed, and with it the thing the friction was building.
Kay’s framework predicts that the consequences of this removal will not be immediately visible. The outputs remain good. The metrics are favorable. The short-term productivity gains are real. The degradation occurs in the trajectory: in the capability the user fails to develop, in the understanding that is not constructed, in the institutional brittleness that accumulates when an organization can produce excellent results without comprehending them. The imagination-to-understanding ratio—Kay’s companion to Segal’s imagination-to-artifact ratio—is the metric that captures what constructive friction was building and what its removal costs.
Keith Sawyer’s research on the conditions for group flow identifies the potential for failure as one of the ten necessary conditions for genuine creative emergence. Groups that cannot fail—whose outcomes are guaranteed—do not produce the intensity of attention that creative work requires. This is constructive friction at the ensemble level: the resistance of genuine risk. AI collaboration removes this friction from one side of the partnership, placing the full weight of caring, risking, and failing on the human.
The pedagogical roots of constructive friction lie in Jean Piaget’s constructivism, developed across decades of research on how children build cognitive structures through active engagement with their environment. Piaget identified a process he called equilibration: when a child’s existing cognitive schema encounters information it cannot accommodate, the resulting disequilibrium—the friction—motivates the construction of a new schema adequate to the new information. Learning is not the transmission of information into a passive recipient but the active reorganization of the recipient’s cognitive structure in response to productive disturbance.
Lev Vygotsky’s concept of the zone of proximal development specifies the conditions under which constructive friction is most productive: the task must be beyond what the learner can accomplish independently but achievable with appropriate support. Too little challenge produces no development; too much produces defeat rather than growth. Seymour Papert extended this framework to computing, arguing in Mindstorms (1980) that the Logo programming environment provided children with concrete, manipulable experiences of mathematical concepts through the constructive friction of building programs that did not work on the first try. Kay built explicitly on Papert’s work in designing the Dynabook’s pedagogical logic.
The concept sits in tension with the entire commercial history of computing, which has treated ease of use as the primary design virtue. The ergonomic tradition, developed to reduce unnecessary strain in human-machine interaction, was appropriated by interface designers and extended into a general principle: any friction is bad friction, and the ideal interface is frictionless. Kay’s objection is that this principle, appropriate for operational tasks where the human operates a machine to complete a function, is destructive when applied to cognitive tasks where the human engages with a medium to develop understanding.
Friction as mechanism, not obstacle. The constructivist insight is that the gap between intention and result is not a flaw in the learning environment but its primary mechanism. The child debugging a program is not failing to learn—the debugging is the learning. The student who cannot immediately solve a problem and must develop a new approach is not being obstructed—the obstruction is the pedagogy. An environment that removes this friction removes the mechanism through which understanding is constructed.
The difference between productive and unproductive friction. Not all friction is constructive. Kay distinguished between friction that develops understanding—the difficulty of learning to program, of building a simulation that does not work, of encountering the gap between intention and output—and friction that is merely wasteful: poorly designed interfaces, unnecessary complexity, barriers that consume effort without producing comprehension. The design challenge of a genuine medium is to minimize the second kind while preserving the first. The AI industry has optimized almost entirely for the elimination of the first kind under the mistaken assumption that all friction is the second kind.
Ascending friction and the AI inversion. As AI tools become more capable, the mechanical difficulty of interaction decreases—prompts become easier to write, outputs become more reliable—while the creative and cognitive difficulty of engaging with the tools meaningfully increases. Sawyer’s improvisational discipline framework captures this: the skill required is not prompt engineering but the evaluative judgment to distinguish genuine insight from fluent confabulation—a judgment that becomes harder, not easier, as the outputs become more polished.
Institutional accumulation and brittleness. Kay’s most consequential application of constructive friction is organizational rather than individual. When an organization routes its analytical and creative work through AI systems that produce output without developing the understanding of the people doing the work, the organization accumulates a debt of comprehension. The outputs remain good; the institutional capacity to understand, evaluate, and adapt them degrades. The organization becomes brittle: capable of production but incapable of adaptation when conditions change, because adaptation requires the deep understanding that only sustained engagement with difficult problems produces.