Esha Zaveri

I have a new paper with Luis Rayo on a key, simple question: will AI end careers as we know them? Link below. We all experience AIs usefulness every day: AI writing code, drafting legal memos, and analyzing spreadsheets. AI can already do many of the tasks that young people just ouf of university can do. This is fantastic for productivity, but it poses an existential threat to how we build careers. The traditional career path is essentially an apprenticeship. You start at the bottom, doing low-value, often menial tasks—the "grunt work." You accept lower pay because you are learning from experts. In effect, you are paying for your training with your time and effort. AI disrupts this bargain. If a machine can do the grunt work, the "currency" that juniors use to pay for their training disappears. If AI can do the junior work, why hire the juniors? And if no one hires them, how will they ever become seniors? This is what I have called the "AI-Becker problem" (after the economist Gary Becker, who analyzed the problem of general human capital acquisition). It’s a crucial question facing organizations today: Will the career pipeline survive the age of AI? In "Training in the Age of AI: A Theory of Apprenticeship Viability," Luis Rayo and I model this tension and find the tipping point. When AI enters an organization, it does two things simultaneously: It raises the Floor (Substitution): AI automates entry-level tasks. The "Floor" is the value the firm can get if they just use AI instead of hiring a novice. As AI improves, the floor rises. It raises the Top (Complementarity): AI can also boost the value of a fully trained expert. Think of an experienced architect using AI design tools to do better work faster. This is the "Top." The survival of the apprenticeship depends on a race between the Top and the Floor. As AI improves, both rise. But which one rises faster? We capture this race in a single statistic: the Expertise Leverage Ratio (R). It asks a simple question: How much more valuable is an AI-augmented expert compared to what AI can do alone at entry? R = (Value of Expert with AI) / (Value of AI alone at entry) Our main result is that there is precise threshold value of R (the mathematical constant "e" below which the apprenticeship collapses. Above, if AI alone is not good enough, or if AI increase the value of the expert enough, the career ladder continues. Why this specific threshold? It emerges from the delicate balance of the training relationship. In an optimal apprenticeship, the master "pays" the novice with knowledge, not cash (at least until they graduate). But the novice always has an outside option: they could quit and use the available AI themselves. To keep the novice from quitting, the master must transfer valuable knowledge just fast enough to make staying worthwhile at every moment. The master wants to slow down training to extract more work from the novice; the novice wants to speed it up. The constant e emerges mathematically as the tipping point that allows the master to optimally "stretch" the training duration. When the Gap is Large (R > e): If the gap between the Top (expert value) and the Floor (AI value) is large, the master has a lot of unique knowledge to "sell." They can stretch the training over an optimal, fixed duration. This gives the master enough time to recoup the costs of onboarding and mentoring. The pipeline is stable. When the Gap is Small (R <e): If the gap is small, the master runs out of saleable knowledge quickly—AI already knows too much. The AI floor gets in the way of the optimal training path. The master is forced to start the training right at the AI floor and must speed up the knowledge transfer. The training window compresses. The Collapse: This compression is critical because hiring isn't free. When the training window becomes too short, the master simply cannot earn enough from the junior's work before they graduate to cover the fixed costs of hiring them. At that point, the pipeline collapses. The firm stops hiring juniors and relies solely on AI for entry work. The future of careers depends not just on how much AI automates entry-level tasks (substitution), but crucially on how much it enhances the value of the experts doing the high-level work (complementarity). Substitution at the bottom threatens training. But complementarity at the top can save it.