Evidence I
1842: The Lowell Stretch-Out
A power loom at a New England textile mill. The overhead belt-drive system transmitted power from a central source to each machine.
In 1842, managers at the Lawrence Company’s Mill No. 2 in Lowell, Massachusetts made what looked like a sure bet. Their weavers, mostly young women, each operated two power looms. The mill had just purchased more machines. Give each weaver a third loom. Output rises 50%.
It did not work. A weaver’s real job was not pulling levers. It was watching. Scanning cloth for broken threads. Catching defects before they ruined a full run. Preventing jams that could destroy the loom. A third machine overwhelmed that attention. The mills cut loom speeds by 15% just to hold quality.1 It took over a year of retraining before weavers could run three looms at the original pace.2
The long arc tells the real story. Over the following decades, through sustained investment in training and process redesign, weavers eventually handled eight or more looms. Output per weaver grew roughly 25 times over the handloom baseline. Bessen’s analysis of original mill payroll records found that 62% of those productivity gains came from human skills enabling more machines.2 Not from better hardware.
Evidence II
1890s: The Thirty-Year Wait for Electricity
Factories began installing electric motors in the mid-1890s. The technology was clearly superior to steam. But for thirty years, electrified factories saw almost no increase in output.
The reason: they swapped in an electric motor where the steam engine had been and changed nothing else. Same layouts built around central drive shafts. Same workflows designed for a previous century.
Meaningful gains arrived only in the 1920s, when manufacturers redesigned everything. Single-story factories. Individual motors on each machine. New material flows and entirely new worker roles. Paul David’s research at Stanford found that roughly half of U.S. manufacturing productivity acceleration between 1919 and 1929 came from this organizational transformation.3 The electrical technology had been available for thirty years.
Evidence III
2026: The AI Productivity Gap
2026: AI tools are open. The workflow hasn’t changed.
The pattern is repeating.
Individual gains from AI tools are real. Developers complete specific coding tasks 55% faster with GitHub Copilot.4 Customer service agents resolve 14% more issues per hour.5 Management consultants finish work 25% faster at 40% higher quality.6
None of it is translating to organizational value. A 2025 NBER survey of nearly 6,000 executives across four countries found that 80 to 89% of firms report zero measurable productivity impact from AI.7 U.S. labor productivity rose only 2.8% in 2025.8
The diagnosis is identical to 1842 and 1895. Companies are installing AI where the old process used to be. Changing nothing else.
The Pattern
| Technology | Initial Result | What Unlocked Value | |
|---|---|---|---|
| Lowell, 1842 | Power looms | 15% speed cut, no output gain | Decades of training + redesign |
| Factories, 1895 | Electric motors | 30 years of flat productivity | Full factory + workflow redesign |
| Offices, 2024 | AI assistants | Task gains, zero firm impact | ? |
Every major technology transition follows the same sequence. The machine arrives. Leaders install it inside the old system. Results disappoint. Then someone redesigns the work, the roles, and the organization around what the technology makes possible.
The technology is never the bottleneck. The redesign is.
Sources
- Bessen, J. “More Machines or Better Machines…Or Better Workers?” NBER Conference Paper, 2008. Citing Montgomery, J. A Practical Detail of the Cotton Manufacture (1840), p. 132.
- Bessen, J. “Technology and Learning by Doing.” Research on Innovation, 2003. Primary records: Lawrence Company Mill No. 2 payrolls, Lowell, MA.
- David, P.A. “The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox.” American Economic Review 80, no. 2 (1990): 355–361.
- GitHub/Microsoft. “Research: Quantifying GitHub Copilot’s Impact on Developer Productivity.” GitHub Blog, 2022–ongoing.
- Brynjolfsson, E. et al. “Generative AI at Work.” NBER Working Paper / Quarterly Journal of Economics, 2023.
- Dell’Acqua, F. et al. “Navigating the Jagged Technological Frontier.” Harvard Business School Working Paper, 2023.
- Bessen, J. et al. “The Effects of AI Tools on Employment.” NBER Working Paper #34836, 2025. Survey of ~6,000 executives across US, UK, Germany, Australia.
- U.S. Bureau of Labor Statistics. Nonfarm Business Labor Productivity, 2025 annual data.