Peter McCrory separates rapid model improvement from the labor-market effects visible today. Claude can already perform sophisticated parts of economic research, and business adoption is rising, yet US unemployment remains near full-employment levels even in occupations with substantial observed exposure to AI automation.
His explanation is that jobs are bundles of interdependent tasks, with important weak links that models do not yet complete reliably. Anthropic's usage evidence also suggests that humans expand the range and complexity of work that AI can handle by supplying judgment, context and evaluation, so partial automation can complement a worker instead of eliminating the role.
McCrory says larger disruption may come when AI-native firms reorganize work rather than merely automating tasks inside existing structures. More capable models could narrow the gap between theoretical capability and actual adoption, while the deepest long-run question is whether AI accelerates innovation itself enough to produce unusually fast economic growth. Sponsor messages are omitted.
Watch on YouTube
