The Solow Paradox Returns
In 1987, Robert Solow observed that the computer age could be seen everywhere except in the productivity statistics. Companies were spending billions on information technology; the computers were visibly present on every desk; the macroeconomic productivity numbers were flat. The paradox haunted the industry for a decade. The resolution came from multiple directions: measurement error, the time required for organizational restructuring, the fact that productivity gains from IT appeared only after the reorganization of work around the computer. The AI transition is replaying the pattern. Enterprise AI spending follows a trajectory that mirrors IT spending in the 1980s and 1990s. AI tools are visibly present in workflows. And macroeconomic productivity has not yet reflected the transformative impact the investment levels would predict. Meeker's 2025 report positions this gap within the historical pattern — counseling patience, since the productivity will come as organizations restructure to exploit the new capabilities.
In the AI Story
The historical analogy is instructive but imperfect. The IT revolution produced its productivity gains primarily through automation of routine tasks — calculation, data storage, document formatting. These tasks did not require judgment; their automation freed humans for judgment-intensive work. The mechanism by which productivity gains were realized was the reallocation of human attention from routine to higher-order work.
AI automates not merely routine tasks but tasks previously considered judgment-intensive. Code generation is not routine. Writing is not routine. Strategic synthesis is not routine. When automation expands into judgment-intensive domains, the reallocation path that resolved the IT paradox — move humans from routine to judgment — requires that there exist a higher level of work to which displaced workers can migrate.
Meeker's concept of the computational labor unit quantifies the amplification: one person augmented by AI produces the output of multiple people. The productivity gain is real at the individual level. The distributional question — who captures the gain, and what happens to those whose output has been consolidated — is a question the productivity data alone cannot answer.
The timeline dimension is critical. The original Solow Paradox persisted for approximately a decade before productivity gains materialized — a period during which organizational restructuring, management innovation, and workforce training converted technology adoption into productivity improvement. The AI transition is compressing this timeline, and the compression may widen the real-terms gap between adoption and productivity in ways the original paradox did not involve.
Origin
Solow articulated the original observation in a 1987 New York Times Book Review essay. The paradox was resolved gradually through the 1990s, as productivity growth rates doubled during the late decade. Meeker's 2025 report explicitly invokes the pattern as the historical template for interpreting current AI productivity data.
The invocation carries forward both the pattern's optimism — productivity will eventually arrive — and its caution — the arrival depends on institutional work that must accompany the technology.
Key Ideas
The gap is a pattern, not an anomaly. Every major technology has exhibited a period during which investment exceeds measurable productivity gains.
Organizational restructuring converts technology into productivity. The computer did not make workers more productive; the reorganization of work around the computer did.
AI's timeline is compressed. The institutional adaptation that produced the IT productivity boom took a decade; AI adoption is proceeding faster than institutional adaptation can match.
The distributional question is separate. Whether productivity gains materialize and how those gains are distributed are two different questions, and the historical record on the second is less encouraging than the first.
Patience is not passivity. The productivity gains from IT required deliberate institutional investment; the gains from AI will require the same.
Appears in the Orange Pill Cycle
Further reading
- Robert Solow, New York Times Book Review (July 12, 1987)
- Erik Brynjolfsson and Andrew McAfee, The Second Machine Age (W.W. Norton, 2014)
- Mary Meeker, Trends — Artificial Intelligence (Bond Capital, 2025)