The productivity paradox: why every new tool made us slower
Computers were supposed to free us. The macro numbers say they didn't. Forty years of the productivity paradox, and the one thing that might finally break it.
In 1987, the economist Robert Solow wrote one of the most quoted lines in the history of technology criticism: "You can see the computer age everywhere but in the productivity statistics."
He meant it dryly. Companies had spent two decades buying computers. The macro-productivity numbers refused to move. The name stuck — the Solow paradox, or the productivity paradox — because the problem stuck. For forty years.
The numbers
US business productivity growth ran around 3% a year in the 1960s. After 1980, it dropped under 1.5% and has hovered there ever since, with a brief 1995–2005 internet bump that economists now broadly agree was an exception, not a turn.
The internet, the smartphone, SaaS, the cloud — none of them moved the macro-productivity needle the way the steam engine, electricity, or the assembly line did.
The pattern
Every wave so far did the same thing: it added a tool. A faster way to write the email. A neater place to store the spreadsheet. A new dashboard to check on top of the seven you already check.
The work stayed. The admin compounded.
A February 2026 NBER study of 6,000 executives put hard numbers on the latest round: 78% of companies said they used AI. 80% reported no measurable impact on earnings.
Why this time might be different
The thing that's different about agents is they don't add a tool. They take a job off the queue entirely. Read the inbox, draft the reply, send it, log it, move on. Not assist. Finish.
Where measured, the gains are real — roughly 40% performance lift on repetitive tasks. An insurance company in a recent case study moved from 6.2 to 8.1 claims processed per worker-hour and dropped data-entry time by 40%.
That's the kind of step-change the digital era kept promising and never delivering.
The macro hasn't caught up yet because adoption is still mostly chatbots and copilots — additive tools, not substitute workers. What changes the macro is when software takes entire jobs off the human queue.
That's what we're building.