The firm keeps the whole efficiency gain; the pipeline bill goes to everyone else

A paper by Nolan Lovett of the NATO Special Operations University, published in the journal Human Resource Development Review, argues that individually rational AI adoption can destroy the collective expertise of an entire occupation, a pattern he calls the tragedy of the cognitive commons ^[1]. The mechanism matters more than the label: a firm that replaces entry-level positions with AI captures 100 percent of the efficiency gain, while the cost of eroding the expert pipeline is spread across every organization drawing from the same talent pool ^[3].

That is the accounting Garrett Hardin described in 1968, when each herder adding an animal to shared pasture profits alone and spreads the overgrazing cost across everyone ^[2]. The asymmetry does the work. If one firm books the whole gain and shares the cost with every other employer hiring from the pool, the private incentive to cut is stronger than the collective interest for any pool with more than one employer ^[21].

The pipeline is not decorative. Lovett's case rests on the claim that experienced professionals are produced by years in junior roles, where beginners take on progressively harder tasks, make mistakes, and learn to repair them ^[4]. He identifies two ways AI interrupts that. The first is elimination: junior positions go because the software does the work ^[5]. The second does not show up on a headcount report. Junior staff stay but, with AI assistance, hit productivity levels that once took years of experience to reach, so the cognitive effort that builds deep domain knowledge never happens ^[6].

Then comes what Lovett calls the validation tether: the ability to supervise AI output depends on exactly the domain knowledge that AI use erodes ^[7]. He cites existing research that catching domain-specific errors in plausible-looking output takes more than spotting obvious contradictions, and that surface-level checks are not enough ^[8]. Habit compounds it. Staff who routinely treat AI answers as reliable lose the reflex to interrogate them ^[9], and the settings where juniors once learned to challenge authority and test claims against reality vanish along with the entry-level roles ^[10].

Timing is why nobody gets punished for this in the current planning cycle. Today's experienced professionals were trained 5 to 20 years ago ^[11], and Lovett puts the full effects of entry-level cuts starting in 2023 somewhere between 2030 and 2045 ^[12], a lag of seven to twenty-two years between the decision and the bill ^[22]. He calls the bind the Human Reserve Paradox: organizations need deep expertise held in reserve for validation, crisis management and situations that overwhelm AI systems, but no single organization has enough economic incentive to maintain that reserve on its own ^[13]. Those who do get through will carry shallower expertise, he argues, having spent their careers orchestrating AI rather than doing independent cognitive work ^[14].

Exposure is uneven. Lovett places software engineering, financial analysis and legal research in the highest vulnerability category on task substitutability, light regulation and strong modularity; medicine and engineering get partial cover from stricter regulatory requirements and stronger professional associations, without immunity ^[15][16].

Watch whether professional bodies in the exposed fields begin certifying domain competence separately from AI skills, which is among Lovett's recommendations, alongside AI-free learning environments, phased AI introduction, and a baseline of human performance established before AI is involved ^[17][18]. He does not propose bans or restrictions on AI use ^[19]. The evidence is still thin: the-decoder reports that a summer 2025 study Lovett cites at length found employment declines in AI-affected occupations, especially among young workers ^[20], and describes the economic evidence overall as mixed ^[23].