AI Could Boost US Economy but Squeeze Jobs and Wages, Anthropic Warns

TEXT START: Artificial intelligence could make the US economy substantially richer by 2030 without delivering better pay or greater job security to knowledge workers, according to a new economic outlook from Anthropic.

The Dissection

The article is a controlled disclosure of the central contradiction: AI can expand output while degrading the economic position of the people whose work it replaces. It presents automation, wage stagnation, unemployment, and capital’s rising income share as risks to be managed rather than as the predictable consequence of ownership and competitive mechanics.

Anthropic’s own scenarios already contain the fracture. In the substantial case, AI performs half of knowledge work while wages remain broadly flat. In the extreme case, AI performs nearly all knowledge work, labour’s share falls from 60 per cent to 45.2 per cent, and capital takes the majority. The article reports the corpse forming, then calls the future “not predetermined.”

The Core Fallacy

The core fallacy is treating productive-participation collapse as a distribution problem. Sharing more GDP may preserve consumption, but it does not restore the mass employment-to-wage-to-consumption circuit that defines post-WWII capitalism.

Under the Discontinuity Thesis, the substantial and extreme scenarios map directly onto P1 and P3: cognitive automation becomes dominant and the majority lose access to economically necessary labour. The article gestures toward P2 but refuses to follow it to its conclusion. Businesses and workers do not collectively choose whether to automate when competitors gain lower costs and higher output by doing so. Adoption becomes coercive through competition.

GDP growth is therefore not evidence of social health. It is the potential growth of an economy in which ownership of productive intelligence is increasingly separated from human labour.

Hidden Assumptions

• Displaced knowledge workers can be absorbed into alternative professions at meaningful scale, despite the article acknowledging that retraining and suitable opportunities may not exist.
• Wage pressure can be contained while AI performs an expanding share of economically valuable work.
• Redistribution can occur without confronting concentrated ownership of the systems generating the gains.
• Business and worker “adoption decisions” remain voluntary rather than being forced by competitive survival.
• GDP remains a useful proxy for broad prosperity even as labour’s share, job security, and productive necessity decline.
• Transfers or wider distribution of AI gains would solve the underlying problem. They could preserve demand, but they would not restore productive participation.
• The extreme scenario is merely a distant possibility rather than the terminal direction implied once AI becomes superior across cognitive tasks.

Social Function

Primary classification: ideological anesthetic and transition management, with a substantial component of partial truth and elite self-exoneration.

The partial truth is real: AI can increase output while wages and employment deteriorate. The anesthetic lies in presenting this as a policy challenge of “sharing the gains” instead of a power transition between owners of AI capital and displaced labour. The language of uncertainty distributes responsibility across businesses, workers, and policymakers, allowing the institutions accelerating automation to appear as neutral observers of an external event.

The Verdict

This is not a reassuring economic outlook. It is a polite forecast of labour’s demotion from participant to claimant. The article identifies the mechanism—automation, wage compression, unemployment, and capital capture—then retreats into governance language before naming the structural result: if the extreme or substantial scenario materializes, the post-WWII labour-consumption order is not repaired. It is replaced.