Noah Smith lays out a framework for thinking about work in the AI era. The core insight: AI is taking over tasks. Jobs are holding steady. Employment for prime-age U.S. workers sits near all-time highs. European firms adopting AI report productivity gains but no job cuts. And remember when Geoffrey Hinton predicted radiologists would vanish? Their demand has actually increased.
The pattern shows up clearly in software engineering. A job that meant 'writing code' a year ago now means checking and maintaining AI-generated code. Research from Humlum and Vestergaard on Danish workers found that AI adoption creates new tasks in content generation, AI oversight, and integration, but produces what they call 'precise null effects' on earnings and hours worked. Work transforms. Income stays put.
Smith sees the workforce splitting. Specialists keep jobs where tasks stay bundled tight, like radiology, where reading scans, patient communication, and treatment decisions all belong to one person. Generalists end up learning what AI handles well and filling the gaps. Some people will build small operations using AI agents to multiply output without headcount. Tools like Zapier, Make, and LangChain let tiny teams automate lead qualification, support tickets, and social media management. The stack unbundles what used to require hiring.