Where Does the Work Left by AI Go? Tesler’s Law and the Workforce

Summary

This interactive essay applies Larry Tesler’s Law to generative AI and workplace change: complexity does not disappear when AI handles part of a task; it moves to someone else. Five fictional scenarios examine refunds promised by a chatbot, a report containing a reversed claim, a junior employee who has not learned the underlying analysis, staffing decisions based on incomplete productivity measures, and an insurance fraud flag that requires human judgment. Each scenario offers different choices that place the remaining work and control on workers, employers, or AI providers. The essay argues that drafting speed should not be treated as completed work when checking, correction, training, or appeals remain. It cites several studies to add context. A study of 5,172 customer-support agents found 15% more issues resolved per hour with AI assistance, with the largest benefits among newer and less-skilled agents; experienced, highly skilled agents saw small speed gains and small quality declines. An early METR randomized trial of 16 experienced open-source developers across 246 familiar tasks found tasks took 19% longer when AI tools were allowed, despite developers later estimating a 20% speed gain. METR’s February 2026 follow-up showed signs of faster work with newer tools, but changes in participants and tasks made the estimate unreliable. Danish survey and administrative data found no statistically significant average effect on earnings or recorded hours during the first two years after ChatGPT’s launch, with effects larger than 2% ruled out. An August 2026 analysis of U.S. ADP data found employment among 22–25-year-olds in AI-exposed occupations 19% below a comparison level, mainly because of reduced hiring; the authors called this an early indicator and did not establish causation or widespread displacement. The essay also notes legal cases involving fabricated legal citations and misleading chatbot fare advice. Its conclusion is that employers should budget for review and training, involve workers in redesigning jobs, and expect providers to make sources easier to inspect. The five scenarios are illustrative, not empirical evidence.