The impact of artificial intelligence (AI) on productivity, employment, and growth is an area of tremendous interest. Unfortunately, it is also one that is cloaked by an absence of data. So Eakinomics read with interest the summary of a new National Bureau of Economics working paper entitled “Firm Data on AI.” The summary notes:

Research teams at the Federal Reserve Bank of Atlanta, the Bank of England, the Deutsche Bundesbank, and Macquarie University fielded identical survey questions between November 2025 and January 2026, yielding responses from nearly 6,000 CEOs, CFOs, and senior finance managers.

What did they find? Some key results of the four-country survey are:

• 69 percent of businesses report using AI, with the fraction ranging from 59 percent in Australia to 78 percent in the U.S. Over the next three years 75 percent expect to use AI.
• Higher-paying firms are more likely to use AI (the incentive to substitute for labor is stronger) and older firms and firms with older directors were less likely.
• The most common use (41 percent of firms) is large language models to generate text.

Of central interest is the employment impact of AI:

More than 90 percent of executives report no effect of AI use on employment over the past three years, and 89 percent report no impact on labor productivity. Among the minority that do report effects, the results tilt slightly positive for productivity, with an estimated average boost of 0.29 percent.

However, executives anticipate much larger effects going forward. They forecast that over the next three years, AI will increase labor productivity by 1.4 percent on average while reducing employment by 0.7 percent, implying a net output gain of roughly 0.8 percent. About two-thirds of the anticipated workforce reduction is expected to come from reduced hiring.

These effects are summarized in this graph (reproduced from the summary):

Obviously, this is not the final word on AI. But to Eakinomics’ eye, the widespread use of AI is the biggest surprise. Based in part on past AAF research, Eakinomics expected relatively small employment impacts, but because of relatively small take-up of AI. This has the former, but not the latter.