Economist Thread: AI Growth Bottleneck Arguments Hold Up

On October 25, economist Afinetheorem posted an 8-tweet thread revisiting the "AI growth bottleneck" thesis: more than two years ago, while organizing economists at an AI lab, he predicted AI would add only about 1 percentage point to growth and was mocked for it. Today, despite rapid technological change, economic growth has barely accelerated—vindicating the economists' call.

Confirmed

• Thought experiment: suppose a $10,000 bricklaying robot is invented tomorrow—its contribution to growth is nearly zero. It doesn't affect brick prices (it may even raise them short-term) or the cost of trucking the machine to the site; real gains only come once architects and designers redesign around the new machine, and historically architectural paradigm shifts take decades—even assuming no technical bottlenecks and huge cost advantages
• Organizational friction: citing "Messy Jobs" by @lugaricano et al., noting that tasks are components of jobs, and jobs often can't be unbundled; incentive structures, career paths, and training systems are all tied to pre-AI-era technology
• On the optimistic "Cursor is spreading fast" narrative, the author says it's only "somewhat" true: most software still isn't written by agents, productivity gains are hard to observe, and no software category has seen prices collapse
• Prediction is hard: a McKinsey 2019 report forecast that 25-40% of jobs would be automated away by 2030, completely missing the arrival of LLM-based AI
• The thread closes by stressing that economics isn't just talk: there are organizational-economics models of restructuring, models of inefficient diffusion/complementary invention, and labor-transition models backed by historical data and empirical work
• The author adds: even if AI/robotics only adds 1 percentage point—let alone 2—of extra annual growth over the next decade, that would be a boom this generation of the West has never seen, hardly a disappointing outcome

Why it matters

The thread offers a systematic counterpoint to today's AI-optimist narrative: multiple bottlenecks—organizational, regulatory, distributional, and paradigm-transition—sit between technical capability and macro growth. When assessing AI's impact, one should distinguish between "successful tech demos" and "realized economic dividends," with the latter often lagging years or even decades.

2026-10-25 ~ 2026-10-25 · 9 related posts

• [source] Economist: skeptics were right about AI growth, bottlenecks still misunderstood — Afinetheorem · 2026-10-25
• Economist: Even a $10k bricklaying robot would barely move the growth needle — Afinetheorem · 2026-10-25
• Thought experiment: a $10k bricklaying robot would barely move growth — Afinetheorem · 2026-10-25
• Architectural paradigm shifts take decades, even with cheap capable robots — Afinetheorem · 2026-10-25
• Tasks live inside unbreakable jobs: organizational friction slows AI — Afinetheorem · 2026-10-25
• "Cursor spread fast" is overblown: most software isn't agent-written yet — Afinetheorem · 2026-10-25
• [source] McKinsey's 2019 forecast of 25-40% job displacement by 2030 won't happen — Afinetheorem · 2026-10-25
• Economists aren't bloviating: org econ models already cover AI friction — Afinetheorem · 2026-10-25
• [source] Even +1-2pp annual growth would be a once-in-a-lifetime boom, economist says — Afinetheorem · 2026-10-25