A new paper in Human Resource Development Review puts a name to a slow-motion crisis: the tragedy of the cognitive commons. Nolan Lovett, a researcher at NATO Special Operations University, argues that when companies replace entry-level workers with AI, they pocket all the efficiency gains while the erosion of professional expertise gets distributed across every employer drawing from the same talent pool. Nobody has enough incentive to train the next generation of experts. And yet everyone needs them.
Lovett identifies two mechanisms doing the damage. The first is straightforward elimination—AI systems absorb the work that once went to juniors. The second is sneakier. Even when entry-level jobs survive, AI-assisted juniors now hit productivity levels that used to require years of experience. That sounds great until you realize the cognitive struggle was the point. Deep domain knowledge doesn’t form when a model does the heavy lifting. Lovett calls the result the “validation tether”: the ability to catch AI’s mistakes depends on exactly the kind of expertise that AI use is eroding. Surface-level checks won’t catch domain-specific errors hiding in plausible-looking output, and people who habitually trust AI answers lose the reflex to question them.
The timeline makes this worse. Professionals working today were trained 5 to 20 years ago. Entry-level cuts that started in 2023 won’t fully reveal their consequences until somewhere between 2030 and 2045. Lovett dubs this the “Human Reserve Paradox”—organizations need deep expertise in reserve for crises and validation, but no single organization has the economic incentive to build it. Software engineering, financial analysis, and legal research sit in the highest vulnerability category thanks to high task substitutability and light regulation. Medicine and engineering get some protection from licensing requirements, but they’re not immune. The full damage, in other words, is already baked in. We just won’t see it for another decade.
Lovett’s prescriptions are notably moderate. No AI bans. Instead, he wants AI-free learning environments, phased introduction of the technology, and baseline human competence before AI gets involved. Professional associations should test domain knowledge through certifications alongside AI skills. The economic evidence remains mixed—a summer 2025 study showed employment declines in AI-affected occupations hitting young workers hardest, while experienced workers held steady or grew. A Federal Reserve Board study found programming job growth has nearly halved since ChatGPT launched. But causal links are hard to prove when monetary policy and post-pandemic overhiring muddy the data. The question isn’t whether AI can do entry-level work. It clearly can. The question is who will be left who understands the work well enough to know when the AI is wrong.
