The real AI jobs apocalypse - Brian Merchant | Substack

TEXT START: AI executives like Anthropic’s Dario Amodei and OpenAI’s Sam Altman have been forecasting an incipient AI jobs apocalypse for three years now.

The Dissection

This is a labor-political polemic presented as a jobs-apocalypse update. It reframes mass replacement as present workplace misery, documents backlash, and offers bans, regulation, and human-centered norms as available countermeasures.

The Waymo handler, human instructors, licensed therapists, and human cartoonists are presented as evidence of human indispensability. Under the Discontinuity Thesis, they are mostly evidence of transition scaffolding, protected enclaves, and maintenance labor surrounding automated capital.

The Core Fallacy

The text mistakes delayed visible substitution for failed substitution. AI making workers miserable is not an alternative to automation; it can be the pre-terminal phase. Employers first use AI for surveillance, speedup, deskilling, hiring avoidance, and wage pressure. Direct elimination comes later, once systems become reliable and institutions adapt.

New York’s school ban and California’s bills are lag defenses. They may slow deployment in selected domains, but they do not refute Cognitive Automation Dominance or Coordination Impossibility. The Waymo handler is not a counterexample to automation. The handler is the human labor required to keep an automated system operational—a servitor or maintenance niche, not a restoration of mass productive participation.

Hidden Assumptions

• Worker anger can block deployment at scale.
• Local regulation can withstand competitive pressure and jurisdictional arbitrage indefinitely.
• Human provision is economically sustainable rather than legally protected, subsidized, or prestige-priced.
• Cultural preference for human creativity can overcome distribution systems optimized for cheap automated output.
• Current visible employment levels measure the real impact, while hiring freezes, task compression, wage suppression, and ownership concentration remain undercounted.
• A role that AI cannot fully replace today will remain economically necessary tomorrow.

Social Function

Classification: partial truth, transition management, and ideological anesthetic.

The article accurately records the first symptoms: worker degradation, public disgust, regulatory resistance, and the hidden human labor propping up automated systems. But it converts a structural transformation into a campaign for better boundaries. It gives readers a language of resistance while implying that bans, norms, and public anger can preserve the employment-centered order.

That is the sedative. The text identifies the pain but overestimates the strength of the painkiller.

The Verdict

This article is right about the immediate form of the crisis and wrong about its systemic meaning. The jobs apocalypse does not need to begin with tens of millions suddenly unemployed. It can begin with miserable workers, fewer hires, intensified surveillance, compressed tasks, precarious maintenance roles, and human work preserved only where law or customer preference temporarily demands it.

The bans and bills can delay the process. They cannot restore the mass employment → wage → consumption circuit if P1, P2, and P3 hold. The article documents the smoke and mistakes it for a firebreak. Under the Discontinuity Thesis, it is an early report from inside the furnace.