AI, Job Loss, and the Bigger Picture

Lately I’ve been thinking a lot about the impact of AI on job losses — especially after seeing some negative reactions to the Ollie and Louie books from people upset that we’ve used AI art rather than hiring illustrators. Those concerns are real, and they echo a much larger debate: what happens to human work as AI becomes more capable?

Economist Pascual Restrepo tackles this head-on in his paper We Won’t Be Missed: Work and Growth in the Era of AGI. His framework starts from a stark but plausible assumption: if AI eventually becomes good enough to do essentially all economically valuable work, then human labor stops being the engine of growth. Instead, growth is driven by how much compute power society can bring online.

Restrepo distinguishes between “bottleneck work” — the essential tasks that must scale for the economy to keep growing — and “accessory work”, which is valuable but not critical for growth. In the long run, bottleneck work gets fully automated, while some accessory work may linger with humans (art, culture, care work, etc.) but mostly as choice rather than necessity.

The unsettling implication is that wages eventually converge to the cost of running the equivalent computation. Labor’s share of GDP shrinks toward zero, and those who own compute infrastructure capture most of the income. Along the way, some workers might enjoy temporary windfalls if their jobs are “last to be automated,” but those gains won’t last.

None of this means humans won’t matter — just that our role in driving economic output diminishes. That raises urgent questions: How do we share the gains of AI? What new safety nets or systems of meaning will we build when much of what we call “work” is optional rather than essential?

This is why I’m excited that the Economics of Transformative AI (ETAI) course — developed at Stanford’s Digital Economy Lab — has been open-sourced. It’s a rigorous, accessible way to think through these issues: from growth models and automation to inequality, existential risk, and governance.

👉 Explore the ETAI course materials (open source)

If you’re curious about the macro-scale human and economic impacts of AI, this is one of the best places to start.