OpenAI Essay Argues Superintelligence’s Greatest Value Lies in Bureaucracy, Not Pure Genius

The Execution Bottleneck in the Age of AGI

OpenAI has launched "The Next Economy," a series of essays focused on the macroeconomic realities of artificial general intelligence (AGI), with an opening paper titled "The eternal complement." Authored by Hemanth Asirvatham and Elliott Mokski, the essay posits that the most transformative contribution of future superintelligent machines will not be generating brilliant insights, but shouldering the monotonous institutional and logistical overhead required to realize them.

The paper, published as part of OpenAI's platform for independent perspectives on AGI, clarifies that the views represent the authors rather than the company's official corporate stance. Asirvatham and Mokski frame intelligence not in isolation, but as one input within an economic production function that is dependent on physical and organizational "complements." While modern generative AI currently lowers the barriers to execution by writing code, drafting prototypes, and synthesizing literature, the authors argue that future frontier models will generate more scientific hypotheses and product designs than the physical world has the capacity to absorb.

Depth Versus Width: The Rising Cost of Empirical Progress

To illustrate the growing friction between thought and reality, the authors point to the shifting economics of discovery. Citing research by economist Nick Bloom and coauthors, the paper notes that sustaining Moore’s law requires more than 18 times as many researchers today as it did in the early 1970s. Across the broader economy, effective research effort has expanded 23-fold since the 1930s, even as measured productivity per unit of research has plunged by a factor of 41. Infrastructure demands have followed suit: specialized equipment usage in science has doubled over the past four decades, the technician workforce is expanding twice as fast as the scientist base, and cutting-edge semiconductor fabrication plants now cost five times what they did 30 years ago.

Contrasting Galileo’s handheld, two-lens telescope with the James Webb Space Telescope—a $10 billion project spanning 300 organizations across 14 countries with 18 mirrors calibrated to a 50-nanometer precision—the essay outlines two potential trajectories for a machine-driven civilization:

• A Civilization of Depth: In this model, machine superintelligence minimizes empirical drag. Rather than building massive new apparatuses, algorithms lean on pure reasoning, high-fidelity computer simulations, and re-analyzing existing datasets. Much like Dmitri Mendeleev predicting unobserved elements via the periodic table or theoretical physicists anticipating the Higgs boson before building particle accelerators, AI would design only the most surgically necessary physical experiments, keeping infrastructure footprints small.
• A Civilization of Width: In this scenario, real-world complexity overwhelms pure calculation. Fields like biology already reflect this constraint: despite in silico computational screening, novel therapeutics must still undergo lengthy, real-world clinical trials on human subjects. Under this model, discovery demands physical expansion, raw materials, automated factories, and massive energy deployment. If machines unlock concepts as grand as Dyson spheres, the vast majority of machine intelligence will not be spent drafting blueprints, but managing the repetitive, astronomical logistics of assembly and supply chains.

Redefining Human Curiosity Amid Machine Scaling

The authors assert that if physical empiricism remains the governing bottleneck, artificial intelligence will increasingly function as an automated bureaucracy—an "institutional intelligence" coordinating global labor and materials under the hard speed limits of chemistry, biology, and the speed of light.

Under such conditions, human workers may retain a comparative advantage in creative ideation and taste rather than routine management, maintaining distinct value by setting research agendas and directing small fractions of machine capacity. The essay marks the latest entry in OpenAI’s broader editorial initiatives around the "Intelligence Age," which have included policy papers on national science agendas and academic publications assessing AI-driven breakthroughs in mathematics and theoretical computer science.