Marx's Machinery Question

In the AI Story

Marx distinguished between machinery in general (any tool extending human capability) and machinery under capitalist relations of production (tools owned by one class and operated by another for the owner's benefit). The distinction matters because it identifies the locus of analysis: not the technical properties of the tool but the institutional arrangements governing its ownership and use. Under different property relations—cooperative ownership, public ownership, individual artisan ownership—the same technical device produces different social outcomes. The power loom in a worker cooperative extends the workers' capacity; the power loom in a capitalist factory extends the capitalist's control. Marx's error was predicting that workers would recognize this distinction and organize revolutionary opposition. His insight was that the distinction is real and that ignoring it produces analysis missing the most important dimension of technological change.

The AI platform economy exhibits the ownership concentration Marx's framework predicts with painful fidelity. The infrastructure enabling the AI artisan—the trained models, the compute capacity, the data pipelines—is owned by Anthropic, OpenAI, Google, Meta, and a small number of competitors. The developers, designers, writers, and analysts using these platforms are not employees (in most cases) but they are dependent—on access they do not control, on pricing they do not set, on terms of service they cannot negotiate. The dependence is not immediately visible because the platforms present themselves as enablers rather than employers, as tools rather than infrastructure. But the structural analysis reveals the relationship: when one party controls access to the means of cognitive production and another party depends on that access to make a living, a power asymmetry exists regardless of the contractual form. The asymmetry is less severe than the factory owner–worker relationship (the developer can switch platforms, negotiate collectively, build on open-source alternatives) but it is present and growing as AI capability concentrates in a smaller number of frontier models whose development costs exceed what all but the largest companies can sustain.

Heilbroner's Marxist analysis reveals two AI-era patterns the triumphalist narrative obscures. First, accumulation dynamics: the companies capturing the largest share of AI's productivity gains are those positioned at infrastructure bottlenecks—cloud providers (Amazon, Microsoft, Google), chip manufacturers (Nvidia), and the AI platform companies themselves. The developers whose productivity has increased twentyfold see some benefit (potentially higher wages, entrepreneurial opportunities) but the surplus—the gap between the value created and the value captured by the creator—flows disproportionately to infrastructure owners, exactly as Marx predicted. Second, dependency deepening: as developers' workflows integrate AI more completely, switching costs rise and bargaining power declines. The developer whose entire practice is built around Claude-specific patterns faces higher costs to switch to an alternative than the developer who learned to code without AI assistance. The dependence that begins as convenience calcifies into structural necessity, and the platform's pricing power increases accordingly.

Origin

Marx's machinery analysis is distributed across Capital Volume I, Part IV ('Production of Relative Surplus-Value'), particularly Chapters 13–15, and in the fragmentary 'Fragment on Machines' from the Grundrisse notebooks (1857–58). Heilbroner's interpretation, developed in Marxism: For and Against (1980) and The Nature and Logic of Capitalism (1985), extracted the analytical core—technology embodies and reinforces power relations—from the revolutionary teleology Marx embedded it within, making the insight available to readers who did not share Marx's conviction that capitalism's contradictions would necessarily produce its overthrow.

Key Ideas

Machinery embodies social relations. Every machine contains implicit decisions about the distribution of power, control, and benefit—decisions made by those who design and own the machinery, not by those who operate it.

Ownership determines outcomes. The same technical device produces different social consequences under different property regimes—cooperative, public, or private ownership generates different distributions of power, autonomy, and surplus capture.

AI concentrates infrastructure control. The platforms enabling the AI artisan are owned by a small number of companies whose market power and switching costs create a structural dependence that resembles—without exactly replicating—the factory owner's power over the factory worker.

Dependency deepens with integration. As workflows integrate AI more completely, the costs of switching platforms or reverting to pre-AI methods rise, reducing the user's bargaining power and increasing the platform's capacity to extract value through pricing, terms changes, or capability restrictions.