Jobs: displacement vs augmentation
Which tasks improve with AI, and who receives the gains? · updated 2026-09-07
Task productivity, job exposure and actual job losses are different measures. AI can help one group while making another workflow slower. Whether saved time becomes better services, fewer roles or new work depends on demand and management decisions as well as model capability.
The concern
Even when average output improves, workers can lose bargaining power or the entry-level tasks through which they learn. Cutting roles before checking errors and mentoring can transfer costs to customers and remaining staff. Company efficiency does not guarantee a benefit for each worker.
The counter-view
The support-agent study found larger gains for less experienced workers, suggesting a route to spreading expertise. AI may remove repetitive work and make new services viable. Results depend on workflow design, supervision and how saved time is used.
The evidence supports variation across people, tasks and tool generations. These studies do not estimate economy-wide net job losses, and WhichAI does not infer such a number from them.