An essay about execution, not genius
On October 1, 2026, OpenAI published "The eternal complement," an essay by Hemanth Asirvatham and Elliott Mokski arguing that the true bottleneck in human progress is not the generation of ideas but the capacity to execute them. The essay is the first entry in a series on "the next economy" hosted on Intelligence Age, a platform OpenAI describes as a home for independent voices exploring an AGI future. The authors write with an explicit disclaimer, and it matters for how the piece should be read.
This analysis separates the essay into three layers: economic findings drawn from published research, forward-looking scenarios that are speculation rather than established fact, and the vendor context in which the essay appears. Where a statistic could not be independently verified against its underlying document during research for this article, it is attributed to the essay authors rather than asserted as confirmed.
The mind-hand mismatch: the essay's opening image
The essay's opening image is astronomical. Humanity's theories describe the universe from its first moments, yet no human has traveled beyond the Moon. The authors call this a mismatch between mind and hand: our minds venture beyond the reach of our hands. They contrast Galileo's telescope, which called on a few dozen hands, with the James Webb Space Telescope, which the essay describes as a ten-billion-dollar observatory built by three hundred organizations across fourteen countries.
This is the essay's core metaphor, and it is a legitimate one. The James Webb Space Telescope primary mirror, with its eighteen hexagonal segments, was assembled in NASA Goddard Space Flight Center's clean room by a large engineering workforce, a fact visible in publicly released NASA photography. The essay's broader point is that reaching farther has historically recruited more of civilization behind each step: "It didn't just take brighter minds to see farther. It took a larger bureaucracy."
This framing is a reasonable description of large-scale science and engineering, though the essay presents it rhetorically rather than as a measured claim.
The statistics: cited, not independently verified
The essay's empirical section cites a series of statistics intended to show that the cost of further progress is rising: sustaining Moore's law requires more than eighteen times as many researchers as in the early 1970s; economy-wide effective research effort rose twenty-three-fold from the 1930s while measured research productivity fell by a factor of forty-one; the technician workforce is growing twice as fast as the scientist workforce; the use of specialized equipment in science has doubled over four decades; and a chip fab today is five times as costly as thirty years ago.
These figures are sourced in the essay to published research, including work by Nick Bloom and coauthors on the declining productivity of research, National Science Foundation NCSES data tables, an SSRN working paper on specialized equipment, and a Congressional Research Service product on semiconductor fabrication costs. During research for this article, the underlying documents at those source addresses could not be retrieved through this newsroom's research tooling, so the specific numbers cannot be confirmed here. They are therefore reported as cited by the essay authors. The general phenomenon they illustrate, that research productivity has declined as research effort has grown, is consistent with the widely cited literature on this topic, but readers should treat the exact multipliers as unverified in this article. If any figure matters to a reader's decision, the underlying sources, listed below, should be checked directly.
The economic logic the essay draws from these statistics is coherent as far as it goes: the economy still advances because a vast increase in research enterprise compensates for declining yield per unit of effort.
Complements and institutional intelligence
From there the essay moves into economics. Two inputs are complements when more of one raises the value of the other; the authors argue that frontier intelligence and the capacity to realize its ideas are complements in exactly this sense. A better telescope makes a good astronomical question more valuable, and a better question makes the telescope more valuable. Ideas, on this view, are fragile things that need institutions, laws, funding mechanisms, supply chains and vast amounts of unglamorous coordination. The authors coin a term for this: "institutional intelligence," the uncelebrated intelligence of execution.
They then pose the essay's central question: where will AI spend its time, on brilliant insight or on institutional competence? And they emphasize that even a civilization of a billion Einsteins would still need most of its Einsteins to mine quarries and manage accounting, because frontier innovation only works when the ordinary world functions around it.
As economic framing, this is a clear and useful way to think about AI's near-term effects. It treats AI as an input to a production process rather than a magic substitute for everything at once. That part is an argument, not a measurement, but it is an internally consistent one.
The access angle: from execution to taste
This is where the essay's access argument emerges, and it is the part most relevant to this desk's coverage. The authors observe that AI is already making execution less scarce: it writes code, searches unfamiliar literature, and turns sketches into working prototypes. Ideas that once required a whole organization can increasingly be pursued by one ambitious person.
They predict this will shift scarcity from execution toward taste, meaning the judgment to decide what is worth making, which question is worth asking, and which direction deserves pursuit. In their framing, the intelligence age begins by giving more individuals the support staff to build ideas of their own, yielding a profusion of idiosyncratic projects outside the institutions that previously acted as gatekeepers.
If this holds, the practical implication is significant: the barriers to building with AI move from capital and organizational capacity toward judgment about what to build. That is an argument with real distributional stakes, favoring broader access to capable AI systems. It should be noted that this argument also happens to support OpenAI's commercial interests, which is exactly why it deserves scrutiny rather than repetition.
The essay also hedges its own claim. The authors warn that if AI begins designing thousands of its own research agendas, ideas could abound faster than infrastructure can absorb them, and society could become more execution-starved than ever. Both halves of this are predictions, not observations. Today's AI demonstrably assists individuals in producing working prototypes and analyzing documents; whether it durably shifts scarcity toward taste is an empirical question that has not been settled.
The two civilizations: clearly labeled speculation
The essay's most speculative section sketches two possible futures. In a "civilization of depth," superintelligence surmounts the need for more physical capital and bureaucratic orchestration: simulations resolve most questions, and only a few targeted real-world experiments remain. In a "civilization of width," the complexity of nature surpasses the ability of any intelligence, human or machine, to make progress without ever-larger real-world experiments. The authors say explicitly that they do not know which pathway, if either, describes the actual future, and that the two are directional possibilities rather than forecasts.
These scenarios rest on contested assumptions: that superintelligence will arrive at all, in some form, and that its capabilities in reasoning-heavy domains such as mathematics will extend to experimental science. Neither assumption is established fact. The essay's own authors acknowledge uncertainty, and this article takes the same position: the two-civilization framing is a thought experiment about how costs of verification might evolve, not a description of what any current system can do. Nothing here should be read as a claim about whether or when transformative AI arrives, or about hard limits on machine intelligence; those questions remain open.
Notably, the essay is less certain than much discourse in this space. It does not promise that intelligence alone dissolves physical constraints. It concedes that even a perfect mind cannot observe the result of an experiment that has not occurred.
What to make of it
The strongest part of "The eternal complement" is its refusal to treat genius as the only input that matters. In public discourse about AI, intelligence is usually the protagonist; execution is treated as an afterthought. The essay's complement framing, and its attention to institutional intelligence, is a genuine contribution to how non-specialists can think about where AI assistance actually helps: not by replacing the whole pipeline from idea to reality, but by cheapening parts of it.
The weakest part is the evidence base. The statistics are secondary citations that could not be verified here, and the essay's future scenarios are unfalsifiable at present. The vendor context is unavoidable: OpenAI benefits from an intellectual climate in which frontier intelligence is the complement whose price is falling. That does not make the argument wrong. It makes it advocacy-adjacent, and it should be labeled as such.
For readers deciding how to act, the essay's most actionable claim is also its most modest one: if AI lowers the cost of turning ideas into prototypes, the scarce skill shifts toward deciding which ideas are worth pursuing. That is a testable proposition, and unlike the two-civilizations scenario, it can be checked against experience over the next few years.