Stay informed with free updates
Just sign up for artificial intelligence myFT Digest — delivered straight to your inbox.
The author is director of the Digital Economy Institute at Stanford University and co-founder of Workhelix.
For more than a decade, economists have wrestled with a modern version of Solow’s paradox. We’ve seen artificial intelligence used everywhere except productivity statistics. Skeptics argue that this is because modern innovations in machine learning systems and current generative AI pale in comparison to the great inventions of the past. But the latest benchmark revisions from the Bureau of Labor Statistics suggest the statistical fog may finally be clearing.
Data released this week provides a notable correction to the view that AI has yet to impact the overall U.S. economy. Initial reports suggested that the steady expansion of the U.S. workforce would continue for another year, but new numbers reveal that total employment growth has been revised downward by about 403,000 jobs. Importantly, this downward revision was made while real GDP remained strong, including a growth rate of 3.7% in the fourth quarter. This decoupling, or maintaining high output with significantly less labor input, is a hallmark of productivity growth.
My own latest analysis suggests that U.S. productivity will rise by about 2.7 percent in 2025. That’s nearly twice the annual average of 1.4 percent that characterized the downturn of the past decade.
This change is consistent with the “J-curve” of productivity that my colleagues and I investigated in previous research. General-purpose technologies, from steam engines to computers, do not provide immediate benefits. Instead, it requires a period of significant and often unmeasured investment in intangible capital, including reorganizing business processes, retraining employees, and developing new business models. At this stage, measured productivity is suppressed as resources are diverted to investment. The latest 2025 US data suggests that we are now moving from this investment phase to the harvest phase, where previous efforts will begin to show tangible results.
Micro-level evidence further supports this structural change. In a study on the employment impact of AI last year, Bharat Chandar, Ruyu Chen, and I identified a cooling in entry-level hiring within sectors exposed to AI. There, hiring of junior talent fell by about 16%, but hiring of talent whose skills were enhanced using AI increased. This suggests that companies are starting to use AI for some codified, entry-level tasks.
Although this trend is suggestive, it requires some caution. Productivity indicators are notoriously unstable, and it will likely take several more periods of sustained growth before a new long-term trend is confirmed. Moreover, strong macroeconomic headwinds, ranging from geopolitical trade wars to fiscal and monetary mismanagement, could impede these efficiency gains.
However, there is reason for further optimism when we distinguish between potential and realized benefits. Many companies use generative AI for only a few tasks. Some employ AI only for translation and summarization. This is the use of what is called a “glorified dictionary.”
Conversely, my company has found that a small number of power users are leveraging interactive conversations with AI agents to automate end-to-end workstreams, such as creating a complete marketing plan, compressing weeks of work into hours. The challenge for companies is not just acquiring technology, but using it to uplift the average employee. This increases not only their own profits but also the productivity of the economy as a whole.
We are moving from an era of AI experimentation to an era of structural utility. We must now focus on understanding how exactly it works. Recovering productivity is not just an indicator of the power of AI. This is a wake-up call to focus on the economic transformation that is coming.
