The J-curve bends: Where AI productivity gains are starting to show up

In a year when U.S. productivity growth roughly doubled its decade-long average, Erik Brynjolfsson sees something most AI commentary has missed: the harvest.

In a recent McKinsey interview, he argues that AI productivity follows a J-curve, the dip-then-rise pattern economists use to describe technology adoption. Companies invest first, get little measurable return for a while, then see gains accelerate as the intangible work (process changes, reskilling, new products) compounds.

The 2025 numbers, in his reading, look like the bend. His own analysis, reported by Fortune, puts U.S. productivity growth at roughly 2.7% in 2025, nearly double the ~1.4% decade average. Q4 GDP was tracking +3.7%. The same year, payrolls were revised down to 181,000 from an initially reported 584,000, a sign that output was rising faster than headcount.

The mechanism named is unglamorous. Brooke Weddle, who leads McKinsey's organizational work, told him roughly 70-80% of the work with client organizations goes into redefining roles, behaviors, skills, and mindsets, not the technology layer. "It's just not the tech or automation," Weddle said. Companies that deploy a large language model, the AI system behind chatbots like ChatGPT, and wait for returns miss the bend entirely.

His "power users" cohort, the small slice of firms automating end-to-end workstreams with AI agents, has collapsed tasks from weeks to hours. That group is still a fraction of the economy.

The macro counterweight is real. Apollo's chief economist Torsten Slok has argued, per Fortune, that AI is "everywhere except in the incoming macroeconomic data," echoing the Solow paradox from the 1980s, named after Nobel laureate Robert Solow's quip that computers showed up everywhere except in productivity statistics. S&P 500 margins and earnings ex-Magnificent 7, the index's other 493 large-cap companies, show no clear AI effect yet. The Fortune piece notes that absence is the principal counterweight to the J-curve reading.

Both readings can be true at once. Capital Economics' Stephen Brown, also cited in Fortune, noted that ICT (information and communications technology) output rose in Q3 2025 even as employment in the sector fell, consistent with AI-driven productivity gains concentrated in specific industries rather than spread across the economy.

That is the J-curve in motion: gains arriving unevenly, by industry and by firm, before they show up in headline GDP.

He is careful about what he is claiming. "Several more periods of sustained growth are needed to confirm a long-term trend," he said in the McKinsey conversation. A geopolitical or monetary shock could offset the early gains. The Stanford senior who opened the conversation with "Is my generation doomed?" did not get a triumphant answer. He got a conditional one.

The conditional matters because the redesign work Weddle describes is not optional. Companies treating AI as a drop-in productivity tool, rather than as a reason to rethink which tasks a role should still own, will keep waiting for the curve to bend. The gains seen are not from AI itself. They come from the organizational changes AI makes newly worthwhile.

The bend is real for some. For most of the economy, it is still a forecast.