AI Should Not Erase Junior Roles: Who Becomes the Experts Then?

AI agents increasingly make it economically attractive to automate entry-level knowledge work. This position paper argues that ML systems should be evaluated and deployed in ways that preserve junior roles as apprenticeship infrastructure, rather than optimized solely for short-term task replacement. The concern is not that AI tools are intrinsically harmful or that junior workers outperform agents; it is that automating formative task portfolios may erode the pathway by which novices become experts. We synthesize evidence across three links: task replacement, where studies report 26–55% productivity gains on tasks often assigned to juniors; role displacement, where labor-market reports show early-career software and technology demand weakening relative to senior roles but do not identify a causal AI share; and expertise erosion, where learning science and early experiments suggest that assisted performance can improve while later unassisted competence may suffer. To make the empirical basis auditable, we provide a coded supplement of 26 public AI-linked workforce restructuring records and a labor-market audit, explicitly distinguishing direct, partial, weak, speculative, and counter-case evidence. The goal of this paper is to treat apprenticeship capacity as part of responsible ML deployment and evaluation.