The boost is always
just around the corner.
Ninety-one confident, dated, named predictions about when information technology and AI will show up in aggregate productivity statistics. Plotted by year said and year targeted. The actual BLS productivity series runs underneath, so the reader can compare what economists said to what the Bureau measured.
Solow's paradox was stated in 1987. It has not been definitively resolved. There was a brief decade, roughly 1995 to 2004, when productivity accelerated and economists declared the paradox solved. Then the acceleration stopped. The predictions resumed.
Everyone agrees the productivity boost is coming. They have for forty years.
In 1987, Robert Solow looked at the rise of the personal computer and the simultaneous flatlining of US productivity growth and wrote, in a brief NYRB book review, "you can see the computer age everywhere but in the productivity statistics." The line became famous. The puzzle it described became known as the Solow Paradox.
It has not been definitively resolved. There was a brief decade, roughly 1995 to 2004, when productivity accelerated and economists declared the paradox solved. Then the acceleration stopped. The predictions resumed. This page collects the confident, dated, named predictions. The same question has been asked for forty years.
Each point is one prediction. The x-axis is the year it was said; the y-axis is the year it targeted. Above the diagonal is the future. Press play to watch predictions accumulate. Hover any point for the quote. Toggle the productivity overlay to see what the BLS actually measured.
Six panels, one per era. The same axes in each. See how the density, direction, and outcome status shift across decades.
The predictions are not random. They cluster in ways that say something about the economics profession itself.
Sell-side analysts are the most consistently optimistic. Goldman Sachs, Morgan Stanley, and JP Morgan have predicted a productivity boost in twelve of the last twenty years. Their average gap between year-said and year-targeted is 4.2 years. When the target year passes without the boost materializing, the next report simply moves the target forward.
Consultancies repeat each other. McKinsey, PwC, and BCG publish similar AI productivity claims annually, often with identical magnitude estimates. McKinsey's 2017 claim of $13 trillion by 2030 and PwC's 2017 claim of $15.7 trillion by 2030 were published three months apart. The similarity is not coordination. It is the same model, run by different firms.
CBO and OMB embed productivity assumptions in fiscal models. These are not predictions in the conventional sense. They are baseline projections required by law. But the productivity assumption is the single most important driver of long-run debt projections. When CBO assumes 1.3% productivity growth, it is also assuming that AI or some other technology will restore the pre-2004 trend. That assumption has been wrong for two decades.
Academic skeptics have been more accurate on timing. Robert Gordon, Daron Acemoglu, and Nicholas Bloom have consistently predicted that productivity gains will be smaller and later than the consensus expects. Their track record on timing is better than the optimists. Their track record on magnitude is harder to score, because "small" is a relative claim.
The productivity mystery is not just about what happened in the data. It is also about what economists said would happen, and when. The Productivity Mystery shows the BLS series, the decomposition, the international comparison, and the seven competing explanations. This archive shows the predictions that preceded each data point.
Together they answer the same question from two directions. The predictions say what people expected. The data says what actually happened. The gap between them is the most interesting thing in macroeconomics.
In macroeconomics, being late by seven years is often the difference between "the prediction failed" and "the prediction was right but the policy response was wrong."
The 1990s IT productivity story is a "right but late" narrative. Brynjolfsson predicted in 1993 that the paradox would resolve by 1998. It resolved in 1995-1996. CBO projected 1.3% productivity growth by 2000. It arrived by 2002. Gates predicted PC productivity gains by 1998. They were visible by 2000.
The McKinsey AI claims of 2017-2019 are beginning to show the same pattern. Sector-level gains in customer service and software development are visible in 2024-2025. The aggregate TFP boost is not. Whether this is "right but late" or "right in the wrong place" is the question the archive will answer in 2030.
Click any row to expand for the full quote, context, and source. Filter and sort using the controls below.
| Predictor | Institution | Year said | Year targeted | Magnitude | Metric | Outcome | Confidence |
|---|
Four candidate explanations for why the productivity boost keeps receding. None of them are mutually exclusive.
A. Diffusion takes longer than capability
A technology can be available for a decade before it shows up in aggregate productivity, because workflows and business processes need to change to accommodate it. Electric motors were invented in the 1880s but did not raise measured factory productivity until the 1920s, after plant layouts were redesigned.
Paul David, "The Dynamo and the Computer," AER P&P, 1990.B. The J-curve
New technologies often reduce measured productivity in the early years as firms invest in reorganization, before paying back later. AI is currently in the investment phase where measured output per hour falls because firms are spending time learning rather than producing.
Brynjolfsson, Rock, and Syverson, "Artificial Intelligence and the Modern Productivity Paradox," NBER, 2017.C. Measurement lags
The BLS productivity series is constructed from data that lags actual production, and does not capture quality improvements, free goods, or unmeasured service-sector output well. The productivity gain may already be there. We may not be measuring it.
Syverson, "Challenges to Mismeasurement Explanations," JEP, 2017; Brynjolfsson et al., "GDP-B," NBER, 2019.D. The skeptic case
Maybe the productivity gains are not there. Not because of measurement, not because of timing, but because the technology is less transformative than it appears. Acemoglu projects ~0.06%/yr TFP gain over ten years. Gordon argues that AI is not comparable to electricity in its economy-wide impact.
Acemoglu, "The Simple Macroeconomics of AI," NBER, 2024; Gordon, "The Rise and Fall of American Growth," 2016.E. Or maybe it is just hard to predict
Hofstadter's Law applied to macroeconomics: it always takes longer than you expect, even when you take into account that productivity always takes longer than you expect. Forty years of confident predictions about "just around the corner" suggests the economics profession has a systematic bias toward near-term optimism.
AGI Horizon collects predictions about intelligence. Productivity Mystery shows what happened in the data. This lab is the bridge between them.
Pre-Paradox & Origin Era (1978-1995)
- Drucker, P. (1978). "The Coming Redesign of Work." Harvard Business Review.
- Bell, D. (1973/1979). The Coming of Post-Industrial Society. Basic Books.
- Solow, R. (1987). "We'd better watch out." New York Review of Books, July 12.
- Brynjolfsson, E. (1993). "The Productivity Paradox of Information Technology." CACM, 36(12).
- Gordon, R. (1993). NBER WP 4549. nber.org ↗
- CBO (1993, 1995). Economic and Budget Outlook. cbo.gov ↗
- Gates, B. (1995). The Road Ahead. Viking.
- Oliner, S. & Sichel, D. (1994). "Computers and Output Growth Revisited." FRB.
The Boom (1995-2004)
- Greenspan, A. (1995-2000). Humphrey-Hawkins testimonies. federalreserve.gov ↗
- Jorgenson, D. & Stiroh, K. (2000). "Raising the Speed Limit." Brookings Papers.
- Brynjolfsson, E. & Hitt, L. (2000). "Beyond Computation." JEP, Winter.
- CEA (1996-2001). Economic Report of the President. govinfo.gov ↗
- McKinsey Global Institute (1997, 2001). Productivity reports. mckinsey.com ↗
The Slowdown (2004-2015)
- Gordon, R. (2012). "Is U.S. Economic Growth Over?" NBER WP 19895. nber.org ↗
- Cowen, T. (2011). The Great Stagnation. Dutton.
- Brynjolfsson, E. & McAfee, A. (2011, 2014). Race Against the Machine / The Second Machine Age.
- Fernald, J. (2014). "Productivity and Potential Output." FRBSF WP 2014-15.
- Bloom, N. et al. (2017/2020). "Are Ideas Getting Harder to Find?" QJE.
AI Prelude (2015-2022)
- McKinsey Global Institute (2017, 2018). AI productivity reports.
- PwC (2017, 2019). "Sizing the Prize." pwc.com ↗
- Acemoglu, D. & Restrepo, P. (2017, 2018, 2019). NBER / AER / JEP.
- Brynjolfsson, E., Rock, D., & Syverson, C. (2017). NBER WP 24001.
- OECD (2018, 2019, 2021). AI and Digital Economy Outlook reports. oecd.org ↗
- BIS (2018). Annual Economic Report. bis.org ↗
ChatGPT Era (2022-2026)
- Goldman Sachs (2023). "Generative AI could raise global GDP by 7%." goldmansachs.com ↗
- Acemoglu, D. (2024). "The Simple Macroeconomics of AI." NBER WP 32487.
- Brynjolfsson, E., Li, D., & Raymond, L. (2023). "Generative AI at Work." NBER WP 31161.
- CBO (2024, 2025). Long-Term Budget Outlook. cbo.gov ↗
- Anthropic (2025). Anthropic Economic Index. anthropic.com ↗
- IMF (2024). World Economic Outlook, April. imf.org ↗
Data
- BLS Labor Productivity (PRS85006152). bls.gov/lpc ↗
- BLS Total Factor Productivity (MPU4900013). bls.gov/mfp ↗
Ninety-one rows assembled across five research batches in May 2026. Sources include NBER working papers, FRB research, CBO/OMB budget documents, OECD publications, BIS annual reports, MGI and PwC reports, Goldman Sachs research notes, academic books, primary press, and congressional testimony. Each row carries a verification level and a source URL where available.
For each prediction we record the year it was said, the year it forecasted, the speaker's role and institution at the time, the concept used (TFP, labor productivity, GDP), the magnitude where specific, and the outcome status. Predictions without a specific year are rendered as horizontal bars. Predictions with ranges use the midpoint.
Brynjolfsson, Rock, Syverson (2017) ↗
Acemoglu (2024) ↗
BLS Labor Productivity ↗
CBO Long-Term Outlook (2024) ↗
Selection bias: this archive collects predictions that named a year. Vague predictions are excluded, biasing the archive toward the brave. Productivity measurement is contested. The Solow Paradox may be partly a measurement problem. Predictor stances change over time. Score predictions individually, not the predictor. The "right but late" framing depends on judgment. Each tag carries an interpretive note.
- Selection bias. This archive collects predictions that named a year. Many predictions are deliberately vague ("AI will eventually raise productivity") and are excluded. The vague predictions are arguably the safer ones.
- Measurement is contested. Service-sector productivity, software quality improvements, and unmeasured digital free goods are hard for the BLS to capture. A "predicted productivity boom that didn't show up" might have shown up in welfare without showing up in measured GDP per hour.
- Most AI predictions are pending. Roughly 60% of post-2022 predictions have target years that haven't yet arrived. The archive will look very different in 2030.
- Predictor stances change. Brynjolfsson has been bullish, then cautious, then bullish again. Score predictions individually, not the predictor.
- Baseline projections are not forecasts. CBO and OMB productivity assumptions are embedded in fiscal models. The genre is contested. Each entry has an interpretive note.