This report was initially featured in TIME’s recently launched AI newsletter, ‘In the Loop’. Readers can subscribe , for free, to receive similar exclusive content directly in their email inbox.
To what extent will artificial intelligence enhance the U.S. economy, if at all?
A recent study conducted by Anthropic, which TIME reviewed exclusively ahead of its public release today, provides at least a partial response to this inquiry.
Through analyzing aggregated data on how individuals utilize Claude during their professional activities, researchers at Anthropic have estimated AI’s potential contribution to annual labor productivity growth—a significant factor in the overall economic expansion—as this technology gains broader adoption.
Their finding indicates that contemporary AI models could boost the U.S. annual labor productivity growth rate by 1.8%, effectively doubling the average growth rate observed since 2019. If labor accounts for 60% of the economy’s total productivity and AI achieves widespread adoption within a decade, the researchers note, “this suggests an overall total factor productivity increase of 1.1% per year.” The study’s authors informed TIME that this figure closely approximates AI’s potential contribution to overall economic expansion. Peter McCrory, Anthropic’s head of economics and a co-author of the study, explained to TIME that, “In these models, labor productivity would typically equate to GDP growth,” provided that labor supply remains constant.
Study Methodology — However, these figures should be viewed with considerable skepticism, as the approach used to derive them is unconventional. Initially, Anthropic researchers developed a tool (referred to as ) enabling them to retrieve specifics regarding actual Claude usage in a manner they assert safeguards privacy. With a dataset of 100,000 conversations, the researchers examined them to ascertain the types of tasks Claude executed in each instance. To quantify the time Claude saved per conversation, they engaged a distinct version of Claude to assess the duration each task would require with and without AI support. Subsequently, they referenced existing economic data to compute the average monetary value of that saved time, contingent on the specific profession. Ultimately, they projected these time savings, weighted by each task’s economic significance, thereby determining the efficiency improvements attributable to AI across various task categories.
Key Limitations — To begin, a primary constraint of the research is its premise that employees dedicate all time saved through AI to additional productive work, as opposed to, for example, devoting more time to family or personal errands. Furthermore, it overlooks the time individuals spend on activities outside their interactions with Claude, such as verifying the factual accuracy of its responses. Another drawback is the researchers’ dependence on Claude for estimating task durations—though they did verify Claude’s assessments against actual data and deemed them satisfactory. Finally, the study does not factor in the rapidly evolving capabilities of AI tools, instead assuming AI’s abilities will remain constant for the upcoming decade. Ceteris paribus, this indicates that the study might be underestimating AI’s potential impact on productivity growth during the next ten years.
Cause for Concern? — The study notably omits any discussion of unemployment, a point particularly striking given that Anthropic’s CEO, Dario Amodei, stated in May that AI could eliminate half of all entry-level white-collar positions within the next one to five years, potentially escalating unemployment to 20%. I inquired with McCrory if the new data supported these concerns. He responded, “In our research, we have not yet specifically investigated the attribution [of the causes of] job displacement, to the extent that it might be happening.” Alex Tamkin, the study’s other co-author, explained that a driving force behind the research was the intention to ready the world for the economic disruptions anticipated with AI. He stated, “Productivity growth benefits the economy, and we are also realistic about the potential effects this technology could have on the labor market. That’s our aim here: simply to introduce more factual information into the discussion.”
