Andrew Yang is back with a prediction that is getting serious traction: AI will wipe out millions of white-collar jobs within the next 12 to 18 months. The Business Insider interview has gone viral, and the claim has split tech observers between "this is obviously true" and "he's been saying this for years."
Both reactions miss the more interesting story.
Set Yang's claim against the CEO productivity data published this week - 6,000 executives telling the NBER that AI has had zero measurable impact on employment or productivity over three years - and you get a genuine analytical puzzle. How can AI eliminate jobs before it shows up in productivity statistics?
The answer, historically, is yes - and it's the answer that makes Yang's timeline worth taking seriously even if his headline number is speculative.
Here's the mechanism that economists call displacement without productivity gain: a company replaces 10 paralegals with 8 paralegals plus an AI legal research tool. The company's measured output stays roughly the same. Its costs fall. Its productivity - measured as output per dollar or output per employee - rises slightly. But the 2 paralegals who lost their jobs aren't counted in the company's productivity report. They're counted in the unemployment statistics. The macro productivity effect takes time to compound and accumulate across thousands of companies. The individual displacement is immediate.
This is why the Solow Paradox framing that economists are using right now is both comforting and alarming simultaneously. Comforting because it suggests AI's transformative effects are real but lagging - just like computing in the 1980s. Alarming because the 1980s and 90s were also a period of significant labor market disruption for workers whose skills became less valuable, even as aggregate productivity eventually improved.
Yang's specific claims center on knowledge work - legal research, financial analysis, coding, content creation, customer service - and he's right that these are the functions where AI has made the most measurable progress. The question isn't whether AI can do useful work in these domains. It demonstrably can. The question is whether companies will restructure their workforces around these capabilities in 12 to 18 months, or whether adoption will continue to lag the way it has for three years.
The honest answer is that Yang's timeline is probably wrong in a specific direction: the disruption will be more uneven and slower in aggregate than his framing suggests, but more severe for specific functions and industries than the zero-impact CEO surveys imply.
What we're likely to see is exactly what always happens when technology disrupts labor: concentration of impact in specific roles, specific industries, specific geographies. A paralegals at a mid-size Chicago law firm faces different pressures than a paralegal at a ten-person firm in rural Ohio. A financial analyst at a hedge fund faces different pressures than one at a regional bank still running legacy systems.
The aggregate number will be reassuringly small. The individual experience will not be. Yang is wrong about the scale. He's right that the question is urgent. And the CEO survey data doesn't actually contradict him - it just describes a different level of analysis.
