The cycle asks what the AI transition means for the human beings inside it, and Lave’s framework supplies the most precise answer to the question the productivity metrics do not capture. When output improves and understanding thins—when developers ship more features while their feel for the systems they are maintaining erodes—what exactly has been lost? Lave’s answer: thick knowledge. Not the propositional facts that a language model can generate, but the contextual, embodied, situated understanding that manifests as professional judgment: the capacity to sense that something is wrong before the formal monitoring catches it, to feel which architectural trade-off matters in this system, to know what the system feels like when it is about to break.
The engineer in Austin whom the cycle profiles—whose output metrics were excellent and who described the experience as hollowness, the difference between being driven somewhere and driving there yourself—is experiencing precisely what Lave’s framework predicts. The sediment is not depositing. The layers of contextual understanding that struggle accumulates are not forming, because the friction that would deposit them has been optimized away. The metrics say she is thriving. The tacit knowledge she is not building will not appear in any quarterly review. It will appear the first time the system breaks in a way nobody expected, and the person who would have felt it coming is no longer there.
Lave’s framework also supplies the most precise instrument for the cycle’s prescriptive question: what institutions, what practices, what deliberate structures can preserve the communities of practice and the developmental trajectories through which thick knowledge develops? The answer is not to reject the tools. It is to insist, as the Liberian tailoring master insisted, on sequencing the apprentice’s access to the practice in a way that ensures each stage of participation deposits a specific layer of contextual understanding. The master made the apprentice press trousers before touching scissors not because pressing trousers is difficult but because handling finished garments first gave the apprentice a felt standard of quality against which every subsequent operation would be measured. Someone must perform the master’s function in the AI age.
Born in 1939 and trained in anthropology at Harvard under John Whiting, Lave conducted her foundational fieldwork in Liberia in the early 1970s, studying tailoring apprenticeships among Vai and Gola craftspeople. What she observed—and what her Western educational training had not prepared her to expect—was that the apprentices learned in the exact reverse of the sequence any curriculum designer would have specified. They began with the final operations, the finishing tasks closest to the completed product, and moved over months and years toward the cutting and fitting that constituted the master’s core expertise. This was not inefficiency. It was a structure of participation that had been evolved, tested, and maintained by generations of practitioners who understood that the apprentice needed to handle the standard of quality before learning any of the operations that produced it.
The Adult Math Project, conducted in Orange County in the early 1980s with grocery shoppers, produced the finding that shattered the transfer assumption: ninety-eight percent accuracy in the supermarket, fifty-nine percent on structurally identical paper tests. The same minds, the same mathematical operations, radically different performance. The variable was context—and context, Lave argued, was not a background condition affecting performance at the margins but the very medium through which intelligence was produced. She formalized this in Cognition in Practice (1988) and with Etienne Wenger in Situated Learning (1991), works that established the framework of situated cognition and legitimate peripheral participation that remains the most rigorous available account of how expertise actually develops.
Situated Cognition. Intelligence is not a property of individual minds extractable in isolation. It is a property of the relationship between a mind and its context—the person-acting-in-a-setting, which includes the physical environment, the social situation, the available tools, the goals in play, and the practitioner’s history of engagement with similar situations. Remove the context and you do not get a slower version of the same intelligence. You get different—and typically poorer—performance, because the intelligence was never solely in the person. The AI model, trained on the propositional residue of millions of contexts stripped of every contextual element, is the purest expression of decontextualized knowledge ever produced.
Thin and Thick Knowledge. Lave’s framework distinguishes between propositional knowledge—context-free, statable, transferable, generatable by a language model—and the contextual, embodied, situated understanding that constitutes genuine expertise. Thin knowledge says what to do. Thick knowledge knows when what to do doesn’t apply, feels which trade-off matters in this specific system, senses that something is wrong before the formal diagnosis confirms it. The distinction is invisible in normal operations and catastrophic in abnormal ones. AI produces thin knowledge with breathtaking efficiency. Thick knowledge is produced only through sustained participation in a practice.
Legitimate Peripheral Participation. Newcomers to a practice do not learn best from simple to complex. They learn best by beginning at the periphery of the practice—performing real tasks with real consequences but limited damage potential—and moving gradually toward its center as their competence develops. The Liberian tailor pressed trousers before he touched scissors, not because pressing is easier than cutting but because pressing is peripheral: errors are recoverable. Each stage deposits a layer of contextual understanding that makes the next stage meaningful. AI disrupts this trajectory with the precision of a surgical instrument that does not know it is cutting a nerve: it makes peripheral tasks unnecessary, allowing the junior developer to skip them, and in doing so it removes the situated engagements through which the trajectory from periphery to center deposits the thick knowledge that central participation requires.
What the Struggle Deposits. The friction of practice is not an obstacle to learning. It is the medium through which learning occurs. Every encounter with a specific problem in a specific context under specific conditions deposits a thin layer of contextual understanding—not the propositional knowledge of what the problem is, but the felt sense of how this kind of system behaves, what this class of failure feels like, what recovery looks like from the inside. Accumulated over years, the layers become the geological bedrock of professional judgment. When AI handles the problem before the practitioner has the encounter, the deposition does not happen. The output is better. The ground is thinner. What the struggle deposits is the knowledge that determines whether a practitioner can be trusted in the situations where formal knowledge runs out.