A joint study by the Wharton School and Accenture covering more than 18,000 tasks and 923 occupations across 19 industries concludes that over 90% of tasks that AI agents could handle are simple enough for less powerful, cheaper models. Between 6% and 11% of agent-capable work generally requires a frontier model. Other 2026 analyses indicate that tokens account for only 20% to 33% of the total cost of operating these systems, with orchestration, human oversight, and integration making up most of the remaining expense. Despite productivity gains, fewer than one-third of organizations report sustained business value at the board level, which could lead executives to scrutinize AI spending more closely.
Source: Read the original article


