Something is showing up in the macroeconomic data that hasn't in over a decade. US labor productivity growth is accelerating. In a February 11, 2026 Wall Street Journal interview Stanford economist Erik Brynjolfsson made a striking claim:
My estimate is that productivity growth was probably about 2.7% in 2025, roughly double what it was during the previous ten years.
2.7% doesn't sound dramatic until you see that the 2014–2018 average was roughly 1.2%. At 2.7%, growth is more than double the trough, a return to pre-slowdown rates not seen since the early 2000s.
The best way to understand productivity is as a ratio of output to work:
Productivity is output divided by workers. If you change the denominator, productivity is much higher than we thought. The fourth quarter is likely to be a pretty good number, and early 2026 is also. We are seeing evidence at the macro level which, until recently, we didn't see.
That's exactly what happened in 2020: COVID wiped out millions of jobs, and the workers who remained were disproportionately in higher-output roles. The 5.5% spike isn't a technology story, it's a composition effect. But look at 2024 and 2025. Employment hasn't collapsed, yet productivity is running near 3%. That suggests something other than workforce composition is driving output higher, and Brynjolfsson is arguing that it's technology.
Now look at the mid-2010s: 0.9%, 1.4%, 0.8%, 1.3%. For half a decade, productivity growth barely broke 1%. That's the trough period, when massive technology investment produced no visible return. The contrast with the last two years is striking.
Brynjolfsson calls this the "J-curve," and he's been studying it for thirty years.
The J-curve is something I've been studying for a long time. In previous technological revolutions, it often played out over literally decades, 20 or 30 years for electricity. The amazing thing is how much faster it's happening this time. We're beginning to see the upturn of the productivity J-curve.
To understand why he's so confident, and why 2.7% isn't a fluke, you need to understand the pattern he's been tracking since the 1990s.
The Productivity Paradox
The story starts with a quote from an economist that might sound familiar to anyone watching the AI debate today:
You can see the computer age everywhere but in the productivity statistics.
This represented the dominant thinking at the time. In the 1970s and 1980s, economists puzzled over a productivity paradox: massive investment in digital technologies wasn't showing up in productivity statistics. Weren't computers supposed to make us more productive?
The J-Curve Pattern
Erik Brynjolfsson spent the next two decades proving they did. His firm-level panel studies analyzing years of Compustat data showed that IT investment increased productivity, but only when paired with complementary organizational changes.
Every time, it played out the same way: a company adopts a new technology, productivity actually gets worse for a while, then slowly recovers, and eventually takes off. Brynjolfsson called it the J-curve:
Anyone who has implemented a new system recognizes this pattern: new technology often hurts before it helps. Learning a new tool or rebuilding a process always slows you down before it speeds you up. What is striking is seeing that same curve in national statistics. And now we're watching it unfold again with AI.
The AI J-Curve
In the 2010s and early 2020s, AI had its own productivity paradox. Despite the incredible advances of machine learning around computer vision, prediction, etc, the macro productivity numbers were low.
This is one of the great puzzles of our era: amazing technologies, but so far, slow productivity growth.
But after tracking tens of thousands of US manufacturers, McElheran, Yang, Kroff & Brynjolfsson's 2025 Census demonstrated the J-curve at the firm level. The study tracked adoption of machine learning and data science systems (what they term "AI") across manufacturing from 2017 to 2021. The firms that recovered fastest were those already digitally mature before adoption.
This whitepaper mainly examines machine learning and data science, things like predictive analytics, computer vision, NLP classification. But as Erik Brynjolfsson noted in that February 2026 interview, we're already into the next wave: agentic AI systems capable of reasoning, planning, and autonomous execution. While Brynjolfsson doesn't make it explicit in the interview, the productivity upturn could be a combination of two overlapping curves: the slower J-curve from 2010s machine learning adoption finally turning upward, and a highly accelerated J-curve from agentic systems that began more recently.
What Brynjolfsson showed in Compustat data in the 1990s, and what the Census now shows in manufacturing plants, may now be visible in aggregate GDP statistics.
Four Lessons for the Mid-Market
They say if you ask ten economists a question you get eleven opinions. I think the same is true for software developers - I have many opinions on how to get the most out of technology so it's fascinating to see what actual comes out of the data. That's why studies like McElheran et al. are so valuable. It adds a layer of objectivity that can only come from studying 10,000 firms. The study identifies four factors that flatten the J-curve for mid-market firms, plus a fifth lesson for multi-location companies.
Research References
| Study | Years Covered | Population | Effect Size |
|---|---|---|---|
| Brynjolfsson & Hitt: Computing Productivity (2003) | 1987–1994 | Large US firms (Compustat panel data) | 5–6% higher productivity; market value of IT capital often 2–3× replacement cost; long-run returns up to 5× short-run estimates |
| Brynjolfsson, Rock & Syverson: The Productivity J-Curve (2019) | 1995–2016 | US economy | General purpose technologies require intangible complements; measured productivity dips then rises |
| McElheran et al.: The Rise of Industrial AI in America (2025) | 2017–2021 | Tens of thousands of US manufacturers (Census data) | Initial productivity drop of −1.33pp; adopters outperform over 4 years; growth strategy, digital maturity, and management discipline flatten the curve |