Can automation raise output while slowing growth?

The paper's central claim is that improving entry-level automation can raise output when it is adopted and still reduce growth and welfare, "even without reducing entry-level employment." The mechanism runs through who works with whom. In the author's task-based overlapping-generations model, novices acquire tacit knowledge by working alongside experts, and "only the most knowledgeable experts manage production and transmit their expertise to multiple novices," which is how best practices diffuse. Automation raises output on impact: routine tasks move from novices to machines, and the novices released by high-skill experts can be reassigned. But when such an improvement "reallocate[s] novices away from the most productive experts," the diffusion of best practices slows and growth falls. The opposite case is stated too: technologies that increase the number of novices learning from the most productive experts "strengthen knowledge transmission and raise growth."

The reasoning rests on incomplete contracts. Tacit knowledge is non-verifiable and embodied in the individual, so it "cannot be pledged as collateral," and training cannot be financed by claims on its future returns. Labor therefore serves as the payment: novices "accept demanding hours and low effective pay to work alongside top practitioners." Because a novice's access to a strong mentor is what is being exchanged, a technology that changes who is matched with whom can change transmission even when headcounts stay the same. The paper's policy reading follows this. It says the need is "to look beyond aggregate entry-level employment and examine how new technologies reallocate junior workers across firms and mentors." The sign of the effect is not fixed. When the CES aggregator parameter η is below 1 and machines are productive enough, "Top experts hire a larger aggregate measure of novices," and the scale effect can outweigh displacement. The paper nonetheless judges displacement "the more salient margin so far in the case of generative AI."

Against the nearest notes, the paper supplies a formal channel for a pattern that the empirical notes describe from the outside. Does generative AI prevent juniors from getting entry-level work? reports interviews in which entry-level software work is redirected before juniors reach it. That paper calls the pathway "unprotected" and treats the loss as a failure of organization, not of GenAI alone. This model says the same gap is structural: because incomplete contracts leave novice training unpriced, a market will not restore it on its own. What makes accountable judgment scarce when AI cognition is cheap? reaches an institutional conclusion from a different direction, listing apprenticeship among the conditions that decide whether AI yields mobility or hierarchy. This paper explains why that condition needs protection. Does AI turn freelance work into validation instead of creation? applies the same channel to freelancers, whose learning comes from paid client work. Does AI assistance help workers learn lasting skills? is individual evidence that assisted gains may not carry over, which is the skill-building worry this paper turns into an aggregate one.

The excerpt does not establish any of this from data. It is a theoretical model, and the excerpt gives no calibration, parameter values or measured outcomes. The employment evidence it describes is cited from other authors, including Hosseini and Lichtinger's "seniority-biased technological change," and the paper itself notes that the available evidence does not yet track "affected workers' subsequent placements." The welfare result also depends on the parameter region: once expert scale is allowed, the effect on long-run growth is "ambiguous." What follows at the strength the model supports is narrower than a forecast. Entry-level employment counts are an incomplete test of AI's effect on expertise, and studies should also record where novices are placed and with whom they work. The paper's policy suggestions (taxes or limits on entry-level automation tools, subsidies for junior positions, apprenticeship-style training) are its own proposals, and the excerpt does not test them.