Stanford's Brynjolfsson sees AI boosting US productivity, but he also co-founded an AI consulting firm

Erik Brynjolfsson, Director of the Stanford Digital Economy Lab, says new economic data shows AI is measurably increasing productivity in the United States. However, the data is noisy, causation is hard to prove, and the GDP growth could stem from massive AI infrastructure spending rather than actual AI productivity gains.

According to updated figures from the Bureau of Labor Statistics, US job growth was revised downward by about 403,000 positions, while GDP growth came in strong at 3.7 percent in the fourth quarter, Brynjolfsson writes in a guest article for the Financial Times.

Fewer workers producing high economic output "is the hallmark of productivity growth," according to Brynjolfsson. He pegs US productivity growth in 2025 at around 2.7 percent, nearly double the 1.4 percent average over the past decade.

Brynjolfsson points to the so-called productivity J-curve, which he laid out in earlier research: Major technologies like the steam engine or the computer don't deliver measurable gains right away. Companies first have to restructure processes, train employees, and build new business models. During that phase, measured productivity actually drops. The latest US data, Brynjolfsson argues, shows that "we are now transitioning out of this investment phase into a harvest phase where those earlier efforts begin to manifest as measurable output."

In his own study from last year, he also claims to have found a roughly 16 percent drop in new hires across AI-related sectors. That finding comes with a big caveat, though: the hiring slowdown could just as easily reflect other factors, like the massive overhiring spree across the tech industry in previous years and the cost-cutting that followed.

Brynjolfsson himself acknowledges that productivity data is noisy and that it takes several quarters to confirm a real trend. It's also worth flagging that Brynjolfsson co-founded Workhelix, a company that helps businesses roll out AI.

Measuring AI's real impact on productivity is still a guessing game

Pinning productivity gains specifically to AI is inherently tricky. At best, macro data shows a correlation with AI adoption, not a causal link. GDP growth is a recognized indicator, but it's an extremely blunt instrument, since it captures overall economic output, not the contribution of any single technology.

And Brynjolfsson's most obvious oversight might be that the GDP growth he cites could largely be driven by the enormous capital being poured into AI infrastructure itself, rather than by productivity gains from actually using AI.

Many companies also don't have reliable ways to measure knowledge worker productivity in the first place. People who write emails, develop strategies, or put together reports often can't be evaluated with simple metrics. Without a solid baseline, there's no way to reliably quantify any improvement from AI. There's just no trustworthy before-and-after comparison.

So for now, much of what's being said about AI-driven productivity gains is still speculative, built on sometimes plausible but fuzzy data. It'll likely stay anecdotal for a while, unless there are waves of mass layoffs that can be clearly traced back to AI.