For a century, productivity has quietly driven improvements in living standards. When productivity growth slows, wages tend to stagnate.
When it accelerates, pay, prices, and the nature of jobs all shift. A new analysis from Anthropic’s Economic Index suggests that artificial intelligence may be poised to accelerate this trend. Researchers examined millions of real-world interactions with Anthropic’s AI assistant Claude, focusing on what the technology actually does in workplaces rather than its theoretical capabilities. The central finding: current AI systems can already handle roughly one-quarter of all tasks in the U.S. economy.
Between 44 and 49 percent of jobs contain some portion of work that these tools can perform. The report estimates that if widely adopted, existing AI could boost labor productivity growth by approximately one to nearly two percentage points annually.
That would potentially double recent productivity trends. Doubling productivity growth means the same number of workers can produce more value in the same amount of time, which allows economies to grow without fueling inflation and enables real wages to rise without harming employers’ bottom lines. However, a significant gap exists between what the technology can do and what people actually ask it to do.
Most current AI capabilities remain unused. The authors suggest that the bigger transformation may come not from new technological breakthroughs but from people gradually adopting the tools that already exist.
This represents a slower, messier, but more likely path than the typical narrative of sudden disruption. When AI is used, the data shows it tends to augment workers rather than replace them. Augmentation means a worker keeps their job but performs it differently—faster, with less grunt work, and with more time for judgment calls.
Replacement means the worker is eliminated. The analysis does not predict this pattern will continue indefinitely, but it reflects what real-world interactions currently show.
While the headline finding—that a quarter of tasks are accessible—might trigger fears of job losses, the data suggests different risks. Businesses may fail to adopt the tools, leaving productivity flat and squandering potential gains. Adoption could happen unevenly, widening the gap between firms that implement the technology and those that do not.
Or the current augmentation pattern could shift toward replacement as the technology improves and companies reorganize work around it. The analysis draws on millions of interactions, a large sample.
However, it represents a single source—interactions with one company’s AI assistant. The patterns may hold broadly or may reflect the specific users who choose Claude. The authors acknowledge this limitation, but the data’s size lends weight to their conclusions.
U.S. productivity growth has averaged roughly one to one and a half percent per year in recent decades. Adding another percentage point or two represents a structural shift that changes calculations on federal debt, retirement savings, and how many nurses a hospital can afford to hire.
The study does not predict certainty about the future. It states that the capacity already exists. Whether the workforce experiences the effects as a gentle lift or a hard shove depends on how quickly adoption spreads and whether the augmentation pattern holds. Those are choices, not inevitabilities.
