Economic impact and productivity modeling of AI agents

Abstract

The spread of artificial intelligence (AI) agents in the economy inevitably asks questions about productivity and the shifts in the labour market and long-run macroeconomic growth. This paper, using a mixed-methods approach, pools microeconomic experimental evidence and macroeconomic modelling, for a thorough assessment of the economic impact of AI agents. The paper reports on four pioneering RCTs that randomized 6,478 workers in writing, coding, customer-service, and management-consulting domains showing a strong skill-levelling effect: Across various industries, overall productivity gains from AI agents range from 14% to 56%, with the majority of these gains occurring at the bottom of the talent distribution and only small or negative effects at the top. The modelling and simulation indicate that macroeconomic benefits will be achieved using a J-curve path and will result from a downward trend in the short term, followed by a long upward trend provided complementary investment is made. Based on macroeconomic forecasts from various organizations, the range of potential potential impacts of the EV transition is wide from +0.5% cumulative GDP growth (Acemoglu, 2024) to +7% cumulative GDP growth over a decade (Goldman Sachs) and +$25.6 trillion value added across industries (McKinsey, 2023). Analysis of the distribution of the labour supply shows that around 19% to 47% of workers are potentially highly affected in terms of the disruption of their tasks, including the relatively high exposure for high-education workers in routine cognitive occupations. The paper provides a comprehensive analytical tool to make sense of micro and macro evidence, unpacks scenarios when AI agents will drive inclusive productivity growth, and outlines a policy roadmap focused on complementary investments, incentives for task-redesign, and workforce transition support measures.