Intelligence is becoming the unit of work

Intelligence never lived only in the skull. I trace how it evolved with our tools, how it just crossed a threshold no individual could cross alone, what machines contribute to it, what judgment refuses to cede, and why I think organizations will soon be measured by it.

Abstract

For a century, work has been organized around information. We produced it, moved it, stored it, sold it. I think that era is coming to an end. What I see coming is an economy organized around intelligence. By intelligence I mean actionable insights, contextually grounded, delivered at the speed and place of decision, at a volume, velocity and veracity that no single human being could ever achieve.¹1This is a reflection, not a report; my argument is philosophical throughout. I want to make that case the only way I know how to make it honestly. I will ask what intelligence has been, watch what it is becoming, and name what it costs to mistake the real thing for its imitation.

I. Thinking outside the skull

I would like to start with a person doing long division on paper. Ask yourself where the thinking is happening. The tempting answer is that it happens in their head, with the paper as a record. Peirce demolished that answer a century and a half ago. Thought happens in signs, he argued, in external, manipulable marks; the paper is doing part of the calculating.²2Peirce (1868): all thought is in signs. Once I accepted that, a pattern opened up across the whole history of the species. Every leap in what we could think has been a leap in what we could think with. The list. The ledger. The diagram. The printed page.

Reading the philosophers of technology, I found they had seen this long before the cognitive scientists made it tractable. Mumford traced industrial civilization to the clock rather than the steam engine, because the clock changed what kind of coordination a mind could lean on.³3Mumford, Technics and Civilization (1934): the clock as the key machine of the industrial age. McLuhan generalized the move: every medium extends some human faculty, and the extension reshapes the faculty it extends.⁴4McLuhan, Understanding Media (1964): media as extensions of man. Arthur closed the loop from the technology side, showing that capabilities evolve combinatorially, each built from prior ones, occasionally capturing a new phenomenon and expanding what the whole system can do.⁵5Arthur, The Nature of Technology (2009): combinatorial evolution. Reading them together, I stopped believing intelligence was ever a fixed endowment. It is a frontier, and we have always drawn it through our instruments. Which leaves a question the philosophy of mind took another century to ask out loud. If the instruments do part of the thinking, where exactly does the mind stop?

II. The mind was never sealed

Clark and Chalmers asked precisely that in 1998, and their answer still strikes me as the most consequential sentence in recent philosophy of mind: the boundary of cognition falls wherever the coupling stops working, and nowhere in particular before that.⁶6Clark and Chalmers, “The Extended Mind” (1998): the parity principle. A notebook consulted as fluently as memory is memory, functionally speaking. Hutchins had described the same dynamic operating at the scale of a crew, finding that the cognition that brings a navy ship into harbor exists in no single head, only in the system of people, instruments, and routines working as one.⁷7Hutchins, Cognition in the Wild (1995).

Clark returned this year with the sequel I had been waiting for. Generative AI, he argues, is simply the newest and most fluid of the non-biological resources we recruit into hybrid thinking systems, and it is our basic nature to build such systems.⁸8Clark, “Extending Minds with Generative AI,” Nature Communications 16 (2025). Wittgenstein settled the deeper point long ago. The meaning of a word lives in how it is used, in the open, in practice, never in some private glow behind the eyes.⁹9Wittgenstein, Philosophical Investigations (1953): meaning as use. I hold the same view of intelligence. It is a performance, visible in what a system does. Simon built my field on exactly that premise. The sciences of the artificial study systems that adapt their behavior toward goals, whether or not the system is a person.¹⁰10Simon, The Sciences of the Artificial (1969), ch. 1. Put the two together and the conclusion becomes hard to escape. If intelligence lives in practice, and practice keeps absorbing new instruments, then intelligence itself can grow. So when I say it is evolving, I mean it literally. This is the next step of a process that began with the first tally mark.

III. The threshold we just crossed

Everything I have argued so far could have been written decades ago, but this next part could not. When I picture what a fully intelligent response to a present circumstance would look like for any organization I have worked with, I see inputs that arrive in many formats at once, documents and sensor streams and conversations and transactions and images, from scattered locations, at different speeds, in an environment that shifts faster than any reporting cycle. If I try to locate the person who could hold all of that in view, I know I could not do it, and neither could the best-briefed executive in history with the best staff in history. The limit is not willingness or talent, but the biological ceiling on how much heterogeneous input one mind can synthesize before the moment to act has passed.

That ceiling is what just moved, and I want to be careful about how I say this. Machine systems can now perform synthesis across types and speeds and deliver the result where the decision is happening, at volumes and velocities without precedent and, when the systems are built honestly, with a veracity that can be audited.¹¹11“The Agentic Organization” (McKinsey, 2025). Floridi gives me the sharpest lens on what kind of machine this is. He conjectures a fundamental trade-off between certainty and scope, with symbolic systems buying provable narrowness and generative systems buying unprovable breadth.¹²12Floridi on the certainty–scope trade-off in symbolic vs. generative AI (2025). The new capability lives exactly on that trade-off, which is why I treat it as a capability and a risk in the same breath. This is what I mean when I say intelligence has evolved beyond humanity. The capability now exceeds anything a human could instantiate, while remaining something only a human can be responsible for. Saying that precisely means splitting the pair apart, so let me take each half in turn.

IV. What the machine brings

I try to be exact about the machine's half, because the whole argument turns on it. What generative and agentic systems contribute is synthesis at speed across types. That is the superpower, and I do not use the word lightly. Epstein once listed sixteen reasons to build a model beyond prediction, among them to explain, to guide what data we collect, to discipline a dialogue, and to reveal the apparently simple as complex.¹³13Epstein, “Why Model?” JASSS (2008): sixteen reasons beyond prediction. When I watch a modern synthesis engine work, I see all sixteen running continuously, over a body of inputs no seminar room could ever hold in common.

Design theory saw where this leaves us earlier than most other fields I read. When a machine can generate and evaluate solutions at scale, Verganti and his colleagues argued, the human contribution migrates from solving problems to framing them, from producing answers to deciding what would count as one.¹⁴14Verganti, Vendraminelli, and Iansiti, “Design in the Age of Artificial Intelligence” (2020). Cain and Pino make the same point for an age of uncertainty. As data and computation saturate design, the scarce skill becomes the navigation between data and decision, the judging of what all this synthesis is for.¹⁵15Cain and Pino, “Navigating Design, Data, and Decision in an Age of Uncertainty,” She Ji (2023). The half I have granted the machine is enormous. The half I am about to keep for us is the reason this essay does not end here.

V. What stays with us

Judgment does not migrate. I hold that view for structural reasons rather than sentimental ones. Suchman showed decades ago that plans cannot capture situated action. The context that makes an action right is produced in the moment, in ways no prior specification can exhaust, which means no synthesis prepared in advance can fully anticipate the situation it lands in.¹⁶16Suchman, Plans and Situated Actions (1987). Geertz taught me the parallel lesson about meaning. Understanding what people are doing requires thick description, the local web of significance they are suspended in, and thin data multiplied a millionfold stays thin.¹⁷17Geertz, The Interpretation of Cultures (1973): thick description.

Star adds the organizational corollary I have watched play out in every engagement I have worked on. Information that travels between communities travels as a boundary object, holding its shape just loosely enough that each side can use it, which means shared data never guarantees shared understanding.¹⁸18Star, “This Is Not a Boundary Object” (2010). And Rittel's designers reason through problems where every formulation already commits to a kind of answer. I know of no synthesis engine for that, because the act is normative. Someone assumes responsibility for one framing over the others.¹⁹19Rittel, “The Reasoning of Designers” (1987). So the division I am defending is this. The machine synthesizes. A person answers for it. Together, and only together, the pair constitutes the intelligence I defined at the start.

VI. The counterfeit

I promised a dialectical argument, and this is where it earns the name, because everything I have described can be faked. Mattern's warning about smart cities generalizes to every product I evaluate. A dashboard is not a mind, and calling an aggregation system "intelligent" is a category error with a sales team.²⁰20Mattern, “A City Is Not a Computer,” Places Journal (2017). Winner explains why I refuse to shrug at the error. Technologies are forms of life; adopting a system that calls itself intelligent restructures who decides, who defers, and what counts as knowing, whether or not anyone intended that.²¹21Winner, “Technologies as Forms of Life,” in The Whale and the Reactor (1986). The values arrive embodied in the system either way.²²22Flanagan, Howe, and Nissenbaum, “Embodying Values in Technology” (2008); Bush, “Women and the Assessment of Technology.”

I can already see the failure modes in the field I work in. Delegating to agents without a matching structure of accountability recreates the oldest problem in organizational economics, the principal whose agent optimizes for something adjacent to the goal.²³23“Rethinking AI Agents: A Principal-Agent Perspective,” California Management Review (2025). And meaningful human oversight, the standard reassurance offered for every such worry, is itself under strain as systems act faster and wider than any overseer can follow.²⁴24“Is Human Oversight to AI Systems Still Possible?” New Biotechnology 85 (2025). So I use a simple test. Where is the framing before the synthesis, and where is the person answerable after it? Show me both and I will call the system intelligent. Remove them and what remains is aggregation wearing intelligence as a costume.

VII. The new unit of work

I saved the organizational claim for last because everything above is required to state it honestly. Schumpeter taught that economic eras are defined by which recombination has become cheap enough to organize around.²⁵25Schumpeter, Capitalism, Socialism and Democracy (1942). By that logic, information's era is ending the way the era of raw materials ended, through abundance. Once synthesis across everything, at speed, in context, is achievable, the scarce thing, the thing worth paying for and building companies around, becomes intelligence in the full sense I have been defending. That is what I mean when I say intelligence is becoming the unit of work.

Organizations already have a name for the capacity this demands, even if they rarely connect it to philosophy. Teece called it dynamic capabilities, by which he meant sensing what is changing, seizing what matters, and transforming what you are.²⁶26Teece, Pisano, and Shuen, “Dynamic Capabilities and Strategic Management” (1997). What the newest literature calls the cognitive enterprise or the agentic organization is, as I read it, that same capacity rebuilt around human-machine pairs.²⁷27“The Rise of the Cognitive Enterprise” (World Economic Forum, 2025). And if my argument holds, no company will ever buy intelligence off a shelf, because intelligence in the full sense includes the framing and the answerability, and those cannot be shipped. What can be built is the organization in which the pair works, where synthesis arrives at the point of decision, where a person owns the frame, and where, of any consequential output, someone can look you in the eye and say I stand behind this. The organizations that learn to produce that, reliably and at scale, are already making the thing everything else will soon be priced against.

Notes

1. This is reflection, not a report. My argument is philosophical throughout.
2. Charles Sanders Peirce, “Some Consequences of Four Incapacities” (1868) and the semiotic writings: all thought is in signs.
3. Lewis Mumford, Technics and Civilization (New York: Harcourt, Brace, 1934), introduction and ch. 1.
4. Marshall McLuhan, Understanding Media: The Extensions of Man (New York: McGraw-Hill, 1964), ch. 1.
5. W. Brian Arthur, The Nature of Technology (New York: Free Press, 2009), chs. 1–2, 6.
6. Andy Clark and David Chalmers, “The Extended Mind,” Analysis 58, no. 1 (1998): 7–19.
7. Edwin Hutchins, Cognition in the Wild (Cambridge, MA: MIT Press, 1995).
8. Andy Clark, “Extending Minds with Generative AI,” Nature Communications 16 (2025), doi:10.1038/s41467-025-59906-9.
9. Ludwig Wittgenstein, Philosophical Investigations (1953); Ray Monk, Ludwig Wittgenstein: The Duty of Genius (1990).
10. Herbert A. Simon, The Sciences of the Artificial (Cambridge, MA: MIT Press, 1969), ch. 1.
11. “The Agentic Organization: Contours of the Next Paradigm for the AI Era” (McKinsey & Company, 2025).
12. Luciano Floridi, “A Conjecture on a Fundamental Trade-Off between Certainty and Scope in Symbolic and Generative AI” (Digital Ethics Center, Yale University, 2025).
13. Joshua M. Epstein, “Why Model?” Journal of Artificial Societies and Social Simulation 11, no. 4 (2008).
14. Roberto Verganti, Luca Vendraminelli, and Marco Iansiti, “Design in the Age of Artificial Intelligence,” Harvard Business School Working Paper 20-091 (2020).
15. John Cain and Zach Pino, “Navigating Design, Data, and Decision in an Age of Uncertainty,” She Ji: The Journal of Design, Economics, and Innovation 9, no. 2 (2023).
16. Lucy Suchman, Plans and Situated Actions (Cambridge: Cambridge University Press, 1987).
17. Clifford Geertz, The Interpretation of Cultures (New York: Basic Books, 1973), ch. 1.
18. Susan Leigh Star, “This Is Not a Boundary Object,” Science, Technology, & Human Values 35, no. 5 (2010): 601–617.
19. Horst W. J. Rittel, “The Reasoning of Designers” (1987).
20. Shannon Mattern, “A City Is Not a Computer,” Places Journal (February 2017).
21. Langdon Winner, “Technologies as Forms of Life,” in The Whale and the Reactor (Chicago: University of Chicago Press, 1986).
22. Mary Flanagan, Daniel C. Howe, and Helen Nissenbaum, “Embodying Values in Technology: Theory and Practice,” in Information Technology and Moral Philosophy (Cambridge University Press, 2008); Corlann Gee Bush, “Women and the Assessment of Technology.”
23. “Rethinking AI Agents: A Principal-Agent Perspective,” California Management Review (2025).
24. “Is Human Oversight to AI Systems Still Possible?” editorial, New Biotechnology 85 (2025): 59–62.
25. Joseph A. Schumpeter, Capitalism, Socialism and Democracy (New York: Harper, 1942).
26. David J. Teece, Gary Pisano, and Amy Shuen, “Dynamic Capabilities and Strategic Management,” Strategic Management Journal 18, no. 7 (1997): 509–533.
27. “The Rise of the Cognitive Enterprise: Why Agentic AI Platforms Are the Next Great Business Revolution” (World Economic Forum, June 2025).

Works Cited