Is AI productivity finally showing up in economic data?

In a Financial Times op-ed, Stanford economist Erik Brynjolfsson argues that the "fog may finally be lifting" on AI's productivity payoff. His evidence is a revision: the Bureau of Labor Statistics cut its reading of 2025 job gains to 181,000, down from an initial print of 584,000 and from 2024's gain of 1.46 million, even as fourth-quarter GDP tracked up 3.7%. Because the economy kept expanding while adding so few workers, Brynjolfsson infers a productivity surge — his own analysis puts 2025 U.S. productivity growth at roughly 2.7%, "nearly double the 1.4% annual average seen over the past decade." Apollo chief economist Torsten Slok is quoted taking the opposite read, quipping that "AI is everywhere except in the incoming macroeconomic data" and noting that employment, productivity and inflation statistics, plus profit margins and earnings forecasts for S&P 500 firms outside the "Magnificent 7," still show no clear AI signature.

Brynjolfsson frames the shift in terms of a J-curve for general-purpose technologies: heavy investment produces no visible benefit at first, then productivity takes off once the investment is absorbed. He says the 2025 data mark a transition "from an era of AI experimentation to one of structural utility," moving "out of this investment phase into a harvest phase where those earlier efforts begin to manifest as measurable output." He adds a mechanism for where the gains are concentrated: not broad-based use, but "a small cohort of power users" who are automating end-to-end workstreams with AI agents and completing tasks in hours instead of weeks. Separately, Capital Economics deputy economist Stephen Brown corroborates the productivity-uptick reading from a different angle, pointing to ICT-sector output rising in the third quarter despite falling employment, and concluding that "AI is making a large contribution to productivity growth."

This sits alongside Brynjolfsson's own earlier, more granular finding that Is generative AI displacing workers at economy-wide scale? — that study locates the labor-market cost of the same transition in entry-level hiring rather than aggregate jobs lost, which is consistent with a macro productivity gain coexisting with a narrower, age-specific hit. It also contrasts with Does AI growth inevitably shift wealth away from workers?, a forward-looking scenario model that predicts labor share falling even as aggregate output and productivity rise; Brynjolfsson's op-ed reports a present-tense productivity uptick without addressing how its gains are distributed between labor and capital. And it offers a macro-level echo of the finer-grained mechanism in When does AI actually boost worker productivity?, which would predict that visible gains cluster among workers already applying familiar skills with AI rather than those learning new ones.

The excerpt does not establish that AI specifically caused the inferred productivity rise. The reasoning is a residual: GDP grew while headcount additions fell, and AI is credited by inference rather than by a decomposition of output growth into AI and non-AI contributions. It rests on one heavily revised jobs report and one economist's own productivity estimate, not an independent or replicated series. Brynjolfsson himself cautions that "several more periods of sustained growth are needed to confirm a long-term trend," and that geopolitical or monetary shocks could offset the apparent gain. The implication the evidence supports is modest: a plausible early signal that the AI investment phase is turning into measurable output for some firms, not a confirmed or broadly distributed productivity takeoff.