Harvard paper records 9% junior employment drop at GenAI adopters within six quarters

The IEEE Spectrum essay describes deliberate insertion of manual control steps in a nuclear plant design to preserve operator skill. This practice addressed the automation paradox where systems hand control back only during anomalies. The same pattern now appears across software, controls, and design roles as models absorb routine tasks that once formed junior expertise. Aviation regulators documented identical erosion in the 1990s when flight management systems reduced hands-on time for new pilots. Harvard and Stanford analyses of payroll and firm data confirm the outcome. Junior employment declined sharply in occupations where models directly substitute entry-level work. Senior employment remained stable or grew. The New York Fed attributes part of the shift to remote work barriers on mentorship, yet both explanations converge on the same fracture: the apprenticeship channel that converts repeated failure into judgment. Safety-critical domains already codified remedies. Nuclear sequences and jet-engine validation retained required manual interventions at fixed intervals. These requirements function as scheduled practice rather than efficiency optimizations. Firms that treat skill maintenance as an explicit design constraint rather than a cost center can replicate the approach in software tooling and review processes. Operational adoption will determine whether the pattern reverses. Organizations that embed mandatory human-only tasks in AI-augmented workflows show slower skill decay in controlled trials. Absent such mandates, the cohort entering the field after 2023 will supervise models without the diagnostic intuition required when those models fail.