The Great Stagnation, and AI as the way out

What "stagnation" means here

The number economists watch is total factor productivity (how much more output we get from the same workers and machines, the part of growth that comes from getting smarter rather than just adding inputs). In the US it grew roughly 2% a year from the late 1940s to the early 1970s, then fell to around 1% or less and mostly stayed there.

Tyler Cowen named it in The Great Stagnation (2011). His explanation: the low-hanging fruit got picked. Cheap energy, mass education and new land each delivered a one-time boost, and nothing since has matched them.

The core ideas

• The easy wins are gone. You can only send a population to school for the first time once.
• Productivity growth slowed across developed economies, not just one country.
• The smartphone era barely moved the numbers. Huge change in daily life, little in measured productivity.
• Bits advanced, atoms didn't. Computing kept improving while energy, transport and construction mostly stalled.
• AI is a candidate general-purpose technology (like electricity or the steam engine, something that changes nearly every industry). But those take 10 to 20 years to show up in the data.
• Technology alone isn't enough. Companies have to redesign processes, retrain people and change management before gains appear. Erik Brynjolfsson calls this the Productivity J-curve: things often look worse before they look better.
• Gains concentrate. Cowen's O-ring point: high-agency people and firms capture most of the benefit, at least at first.

Stagnation is real

• Tyler Cowen: the original thesis. Treats AI as the live test of whether stagnation ends. "Big, but slow." Co-founded the Progress Studies movement with Patrick Collison.
• Robert Gordon, The Rise and Fall of American Growth (2016): the pessimist. The great inventions of 1870 to 1970 were one-time gains that won't repeat.
• Peter Thiel: "We wanted flying cars, instead we got 140 characters." Technology outside computing stalled. The fix is definite optimism: specific plans for specific futures.
• Daron Acemoglu: agrees on the diagnosis but blames market power and bad deployment choices, not complacency. His own estimate of AI's near-term productivity boost is small, around 0.07 percentage points a year. Without better institutions he expects "so-so automation" (his term with Pascual Restrepo for automation that replaces workers without making much more).
• Erik Brynjolfsson: the gains are real but lag adoption while organizations reconfigure. The bottleneck is the complementary changes, not the technology.

AI gets us out, and soon

• Marc Andreessen: stagnation is regulatory, not technological. AI fixes it by making intelligence abundant instead of scarce.
• Dario Amodei: expects a fast exit. "Machines of Loving Grace" (2024) argues powerful AI could compress a century of progress in biology into a decade, and makes specific, testable claims.
• Demis Hassabis: the strongest track record. AlphaFold and DeepMind's other science systems are the most concrete evidence that AI can break stagnation in a specific field.
• Sam Altman: "Moore's Law for Everything." The cost of intelligence keeps falling fast, and superintelligence may be "a few thousand days" away.
• Leopold Aschenbrenner: effective compute grows from both hardware and algorithmic efficiency, each by large factors a year. If AGI arrives around 2027, the exit is fast.
• Peter Diamandis: rejects stagnation entirely. AI, biotech, robotics and energy are already converging.
• Dave Blundin: scaling keeps working, and AI-native delivery reshapes software and knowledge-work economics within a few years.
• Eric Schmidt: conditional yes. It depends on the US winning the capability race and building the energy (multi-gigawatt data centers, nuclear) to run it.

Not so fast

• Daron Acemoglu: modest gains unless institutions change (see above).
• Yann LeCun: human-level AI is further off than optimists claim. Today's language models lack world models, planning and persistent memory.

Other angles

• Benedict Evans: agrees it's big but slow, stays agnostic on the exit. The real question is product fit and adoption timing, not raw capability. Near-term impact gets overestimated, long-term underestimated.
• Salim Ismail: stagnation is an organizational problem. Companies built to use external resources and technology outperform, and AI speeds up that shift.
• Mustafa Suleyman: The Coming Wave implies stagnation ends whether we like it or not. The question is containment, not direction.

The evidence so far

• US productivity growth: about 2% a year in the postwar boom, about 1% or less since.
• Counterpoint: GDP may undercount digital goods that are free or cheap, which Brynjolfsson has written about.
• AlphaFold predicted the structures of more than 200 million proteins, the clearest single example of AI unlocking a stuck scientific field.
• The largest tech companies are spending hundreds of billions of dollars a year on AI data centers, and agent tools are entering real workflows.
• Early studies show 10–40% gains on specific tasks. Those gains don't show up in national productivity data yet.

Open questions

• Does AI compound faster than past general-purpose technologies? Brynjolfsson's J-curve says the gains lag. Aschenbrenner says they accelerate.
• Will AI lift productivity measurably by 2030? Acemoglu says no. Amodei and Altman say dramatically.
• What's the real bottleneck: the technology (LeCun), institutions (Acemoglu), organizations (Ismail) or energy (Schmidt)?
• Does faster science, like protein folding and materials discovery, turn into broad economic growth, or stay inside those fields?
• If Thiel is right that bits advanced while atoms stalled, does AI change the physical world or just speed up the digital one?

All theories · clickbridge.com · Rich Price · last reviewed September 2026 · People linked here have their own pages at people.clickbridge.com