The AI Productivity Paradox: Why the Tech Revolution is Missing from Economic Data

A recent NBER study shows that the vast majority of CEOs see no measurable impact from AI on productivity or employment despite massive financial investments. This trend revives the 1980s 'Solow Paradox,' which describes a lag between technological advancement and economic results. Experts suggest that while current data is disappointing, a productivity surge may eventually occur once companies move past the investment phase into effective implementation.

Key Points

• A study of 6,000 executives found that 90% of firms report no productivity or employment gains from AI over the last three years.
• The current situation mirrors the 1980s 'Solow Paradox,' where massive IT investments failed to yield immediate measurable economic growth.
• There is a stark disconnect between the high volume of AI mentions in corporate earnings calls and the actual time executives spend using the tools.
• Economists are debating whether AI will follow a 'J-curve' trajectory, where productivity gains are delayed until the technology is fully integrated.
• Worker confidence in AI utility is declining, with an 18% drop in trust despite an increase in regular usage.

Sentiment

The community is predominantly skeptical that AI is currently delivering meaningful macro-level productivity gains. There is a strong undercurrent of cynicism about corporate AI hype versus on-the-ground reality, and many commenters view the article's findings as unsurprising validation of their own experience. However, the tone is more 'told you so' than hostile, and developers who use AI for coding form a notable minority defending its practical value. The discussion is substantive and well-engaged rather than dismissive.

In Agreement

• AI-generated 'workslop' creates downstream review burdens that offset any time savings, making the net productivity impact negligible or negative
• Workers who benefit from AI pocket the time savings rather than increasing output — finishing tasks faster but not pulling more work
• Much of the white-collar work being 'optimized' by AI may have no real economic value, echoing Graeber's Bullshit Jobs thesis
• AI errors and hallucinations require verification effort that often equals or exceeds the time saved, particularly in non-coding contexts
• The comparison to early IT adoption is apt — massive investment with costs currently outweighing benefits across the economy
• Reports written faster by AI and summarized by AI create circular, zero-value workflows where no human actually processes the information

Opposed

• AI adoption costs are dramatically lower than early IT — a subscription costs as much as Slack, and onboarding is trivial compared to teaching 1970s workers to use computers
• Coding productivity with AI is genuinely transformative, with latest models understanding full codebase context and implementing nuanced solutions
• Productivity gains are real but invisible to measurement — workers use personal subscriptions, and improvements get attributed to other causes
• GDP is up while hiring is down, suggesting productivity IS increasing even if hard to attribute directly to AI
• We may be in the very early stages equivalent to PC DOS 3.x — the Windows 95 moment for AI hasn't arrived yet