The essay challenges the dominant story told by leading AI labs: that building ever more powerful artificial intelligence will automatically deliver explosive productivity, surging GDP, and widespread prosperity. While this narrative is appealing to those developing and financing advanced models, it overlooks a fundamental economic reality. An economy is not simply about producing more; it depends on circulation. Production must be matched to demand, and demand requires broadly distributed purchasing power. When income and bargaining power collapse for large parts of society, productivity gains alone cannot sustain prosperity.
Drawing on William Blake’s metaphor of “the prolific” and “the devourer,” the text argues that output must ultimately be consumed for an economy to function. Many visions of the AI economy assume that production can continue to scale even as human labor is displaced and wages fall, with consumers somehow continuing to buy. This leads not to abundance, but to an unhealthy system in which profits and capabilities accumulate at the center while the rest of the economy is starved of demand. The result resembles economic “congestive heart failure,” where circulation breaks down.
The essay then distinguishes between two intertwined futures: the discovery economy and the labor replacement economy. AI has enormous potential to accelerate discovery, enabling breakthroughs in energy, materials, and medicine. However, discovery is not the same as economic value, and it certainly does not guarantee shared prosperity. Between discovery and broad benefit lies a long, fragile pipeline of validation, regulation, manufacturing, distribution, and adoption. If AI speeds up discovery without accelerating diffusion, society gets impressive headlines and concentrated wealth, but not broad-based growth. The danger is a tall peak of value creation that never spreads into a wide plateau of shared prosperity within the AI economy.
Control over key choke points determines which path is taken. Ownership of compute, data, models, intellectual property, and distribution channels shapes who benefits from innovation. Tight control and slow regulation can turn breakthroughs into a form of “discovery feudalism,” where progress exists but its benefits remain narrowly captured. By contrast, open standards, interoperability, faster regulatory processes, and multiple routes to market can turn AI-enabled discovery into a generalized engine of progress.
The labor replacement narrative introduces an even sharper constraint: demand. Replacing wages with cheap inference undermines the consumer base of the economy. As the wage share falls, instability rises, social conflict intensifies, and long-term investment weakens. Historical examples, from Henry Ford’s high wages to the infrastructure built around mass car ownership, show that productivity only becomes prosperity when institutions and complements spread purchasing power and create new customers. The same lesson applies to the AI economy today.
The essay argues that decentralized architectures are essential to avoid premature concentration. Historically, decentralization has fueled innovation, while centralization has captured value and slowed progress. The current trend toward large, closed models and concentrated cloud infrastructure risks cutting off circulation by encouraging profit hoarding rather than reinvestment. To avoid this outcome, AI development must prioritize open interfaces, interoperability, and competitive ecosystems.
Ultimately, the choice is stark. We can build an AI economy that concentrates value and hollows out demand, or one that circulates—where discoveries diffuse, productivity dividends become time and income for many, and institutions evolve fast enough to support shared flourishing. Intelligence alone is not enough; prosperity depends on designing the flywheel that keeps the system moving.