Here's the most counterintuitive AI story of the year: thousands of executives, surveyed by the National Bureau of Economic Research, say that despite two-thirds of their companies using AI tools, the technology has had zero measurable impact on employment or productivity over three years.
This is not what the vendors promised. But it's exactly what a 40-year-old economic observation would predict.
In 1987, economist Robert Solow made the observation that became known as the Solow Paradox: "You can see the computer age everywhere but in the productivity statistics." Computers were everywhere. Productivity growth, after briefly spiking, had actually declined - falling from 2.9% annually between 1948 and 1973 to just 1.1% after 1973. The computers that were supposed to transform the economy weren't showing up in the numbers.
History knows how that story ended. The productivity gains came - they just arrived fifteen years late, materializing in the 1990s internet boom as companies finally figured out how to restructure workflows around the new technology. The lag wasn't evidence that computers didn't matter. It was evidence that transformative technology takes time to actually transform anything.
The NBER study of 6,000 executives is now resurrecting that same debate. Apollo's chief economist Torsten Slok captured it precisely: "AI is everywhere except in the incoming macroeconomic data." Companies are paying for Copilot licenses, deploying AI tools, training employees. The productivity needle hasn't moved.
But here's the part that gets complicated: this does not mean AI doesn't matter. It means AI might matter enormously - just not yet, and not in the way the hype cycle suggested.
Stanford economist Erik Brynjolfsson argues for a "J-curve" pattern: initial investments in a transformative technology actually suppress measured productivity before the gains compound. You're spending time and money on the technology itself, on reorganizing workflows around it, on training people to use it. All of that consumes resources before it generates returns.
The analogy that feels most apt: when electricity first became available to factories in the late 1800s, manufacturers initially just replaced their steam engines with electric motors - same equipment, new power source. The productivity revolution came decades later, when factory designers started from scratch and built layouts that only made sense with electric power. The technology's transformative potential required reimagining the entire workflow, not just plugging in a new motor.
AI used as a fancy autocomplete feature in the existing workflow will show up as a minor productivity gain, if at all. AI used to fundamentally restructure which tasks exist and who does them is a different proposition entirely.
The executives surveyed still forecast AI will boost productivity by 1.4% and output by 0.8% in the coming three years. They believe in the technology even though they can't yet measure it working. That's either irrational optimism or hard-won pattern recognition from people who've managed technology investments before.
The honest answer is: we don't know yet. The Solow Paradox resolved eventually. Whether the AI equivalent resolves on a similar timeline, faster, or slower depends on factors that aren't visible in today's data. What's clear is that anyone claiming to know definitively - either that AI is already transforming everything, or that it never will - is selling you something.
