“Generative AI at Work,” NBER Working Paper 31161, was released in April 2023 by Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond. The authors describe it as the first study of generative AI’s effect on productivity in a real workplace rather than a laboratory. They studied 5,179 customer-support agents at a software firm who were gradually given access to a conversational AI assistant built on a GPT model.
According to the paper, access to the AI tool increased the number of issues resolved per hour by 14 percent on average. The gain was highly uneven: the paper reports a roughly 34 percent improvement for novice and low-skilled workers, while the most experienced and highest-skilled agents saw little or no measurable change. The authors interpret this as the AI tool disseminating the implicit best practices of the firm’s most able workers to everyone else, helping newer agents “move down the experience curve” faster.
The study also reported softer effects the authors attribute to AI assistance: improved customer sentiment, higher employee retention, and signs that workers learned from the model over time. The paper is frequently cited alongside the Noy and Zhang writing experiment as early evidence that generative AI tends to compress productivity differences between workers - a pattern Daron Acemoglu cites when discussing why AI’s near-term gains may flow disproportionately to less experienced staff.