What does the research evidence actually say about productivity gains from AI tools in the workplace?

Research evidence on AI tools' productivity gains in the workplace reveals a notable paradox: while individual and task-level improvements are evident, broader firm-wide and economy-wide impacts remain mixed or elusive. Surveys of corporate executives indicate positive labor productivity gains that vary by sector, with over half of firms having invested in AI and expectations for stronger effects in 2026, though perceived benefits often exceed measured outcomes [1]. Government pilots and studies confirm AI boosts worker productivity for specific tasks, such as document drafting or customer service, but these micro-level gains have not translated into statistically significant firm-wide improvements in output per worker or total factor productivity, partly due to adoption challenges and increasing system complexity [2][3][4][5]. At the macro level, there's a disconnect between strong individual productivity evidence and limited aggregate results, with scarce signs of real economy-wide gains despite vast investments [6][8][9][10]. Some sources highlight significant industry-level advancements where AI outpaces human labor, yet this raises concerns about job displacement without clear net productivity booms [7][11]. Managers often report personal productivity boosts, contrasting with workers' experiences, underscoring implementation hurdles [12].