Chaining Tasks, Redefining Work: A Theory of AI Automation

URL SCAN: Chaining Tasks, Redefining Work: A Theory of AI Automation
FIRST LINE: # Economics > General Economics

THE DISSECTION

The paper models AI automation as a workflow problem. In a Leontief sequence, AI adoption spreads through contiguous task chains, so firms may assign even human-advantaged steps to AI when doing so completes a cheaper automated segment. The paper’s real contribution is showing that automation is structurally clustered and can arrive in blocks rather than as a smooth task-by-task replacement.

THE CORE FALLACY

The category error, judged against the Discontinuity Thesis, is confusing firm-level task assignment with economy-wide human viability. The model asks where humans remain inside a production chain. DT asks whether human labor remains necessary to the system, who owns the chain, and whether displaced workers retain purchasing power.

The productivity J-curve is therefore not evidence of durable human protection. It is a lag: returns remain weak while complementary tasks are assembled, then accelerate once the chain closes. A temporary bottleneck is not a reversal. The paper’s own result—that AI can absorb human-superior tasks as collateral—weakens comparative-advantage defenses rather than restoring them.

HIDDEN ASSUMPTIONS

• Demand, investment, and profits remain available to fund chain completion.
• Firms can reorganize workflows without decisive legal, institutional, or political resistance.
• Task adjacency is a sufficient representation of real coordination complexity.
• AI quality, cost, and deployment conditions can be treated as technical parameters rather than ownership and power variables.
• Labor remains a meaningful residual input merely because some steps are not yet automated.
• Correlation between neighboring AI-executed steps is evidence of the model’s mechanism rather than selection, common shocks, or reverse causality.
• The model can remain at the firm level without addressing wages, consumption, concentration, transfers, or mass productive exclusion.

SOCIAL FUNCTION

Partial truth with a transition-management function. It correctly identifies clustering and destroys the comforting claim that AI only takes tasks at which it already has superior comparative advantage. But by stopping at workflow optimization, it permits discussion of where humans fit while avoiding the harder question of whether most humans remain economically necessary at all.

THE VERDICT

This is not a rebuttal to discontinuity. It is a map of the attack surface. The paper refines P1 by showing how AI can seize entire workflow segments, including tasks humans perform better in isolation. It supports P3 by implying that labor displacement can be lumpy and discontinuous once chains become viable. Its missing variables—ownership, distribution, purchasing power, and productive necessity—are not peripheral. They are the regime-level consequences.

The paper explains how the machine eats the workflow. It offers no mechanism that preserves the workers removed from it.