AI Is Cognitive Leverage

There is a concept in finance called leverage. I cannot stop connecting it to what I have been thinking about with AI.

The connection surfaced for me today, while reading a piece by Gabriel Di Lelle on how AI is moving through organizations. His piece is worth reading directly - he frames the dynamics with more granularity than I will here. What it surfaced for me is something I have been trying to put words to from a different angle - an angle that seems to answer a question I left open in Where Advantage Migrates.

In that essay, I argued that as AI commoditizes execution, advantage migrates upward - toward the architecture of judgment governing decisions, rather than the execution itself. What I did not address was the mechanism. How does that migration actually happen?

The frame I keep returning to does not come from technology or organizational design. It comes from finance.

In a strategic finance class I have been taking at Harvard, the concept of leverage has been on my mind. At its core, leverage allows a relatively small base of capital to control or generate disproportionately larger outcomes. If you invest $1M of your own capital and borrow $9M, you control a $10M asset, and a 10% rise produces a gain on the full $10M, not on your original $1M. The structure amplifies returns, and it also amplifies losses. Small advantages compound faster; small errors get magnified. Strategically, leverage changes scale, sensitivity, volatility, and dependency structures.

That structure maps with surprising precision onto AI inside organizations.

AI is cognitive leverage. Not capability, not a tool in the conventional sense, but leverage - applied to whatever cognitive capacity already existed in the person, the team, or the organization using it. A small amount of high-quality reasoning, given AI, now influences much larger systems than it could before. A small amount of weak reasoning, given AI, scales much faster than it ever could before. AI increases the return potential of cognition while simultaneously increasing systemic exposure to poor cognition. Like all leverage, it amplifies both capability and risk.

Layer 1: The user

When I first sat with this frame, I assumed the leverage was operating at a single layer - the user. Each employee, given AI tools, is now operating with cognitive amplification they did not have a year ago. The asymmetry inside that amplification is the first observation worth naming.

A rigorous thinker, given AI, encounters a sparring partner. Every output gets interrogated. Structure gets pulled apart. Premises get tested against experience. Their cognitive capacity does not just operate faster - it compounds. The leverage produces returns.

A more passive thinker, given AI, encounters something different. The output ships faster, the surface looks polished, confidence increases - but the underlying thinking does not deepen. The leverage produces something that looks like productivity, while the cognitive position underneath has not improved. In some cases it has thinned, because the friction that previously forced engagement is gone.

What changes here is not cognitive quality itself - it has, in fact, diverged more than ever. What changes is something more specific: the visible artifact is no longer a reliable proxy for the underlying cognition that produced it.

Pre-AI, better thinkers usually produced visibly better artifacts. Rigor manifested in the work itself - in clearer writing, stronger synthesis, better structure, sharper communication, more polished outputs. Organizations used these as signals of capability, and most of the time the signal was reliable.

With AI, that signal has collapsed. Fluency has become cheap, polish has become democratized, structure has become automated, articulation has become assisted. The exact properties that used to differentiate rigor from fluency are now produced by the tool, not by the thinker. The consequence is structural: two people with the same role, the same level, the same task, and the same AI tools can now produce decks and documents that look similarly sophisticated on the surface, while the underlying reasoning quality remains radically different. The downstream consequences emerge much later - in decision quality, strategic coherence, resilience under pressure, execution integrity. By the time they are visible, the cognitive divergence has been compounding invisibly for months.

The deeper claim is the inversion. Cognitive asymmetry may actually be increasing - precisely as visual asymmetry is disappearing. Two trajectories diverging sharply underneath, two artifacts converging on the surface. (I have written about this dynamic in more depth in AI Doesn't Equalize Thinking. It Amplifies It.)

This is the visible layer of the leverage. It is real, and it is important. But it is also the easier half of the story.

Layer 2: The designer

The user layer is only one half of where the leverage is being applied. There is a second layer, almost entirely invisible, and I think it matters more.

The agents that users are reasoning with are not neutral. Behind each one is a small number of decisions about its master prompt, its knowledge domain, its tone, its role, its assumptions - and those decisions condition how the user reasons inside the conversation. Multiply that by thousands of users, and the cognitive priors of a small number of designers begin to propagate across the organization invisibly.

In an exchange with Gabriel about this, he framed the work this requires beautifully: "a new organizational discipline that does not yet exist - the design, deployment, and updating of conditioners of human reasoning." That phrase has stayed with me. Conditioners of human reasoning. It captures what the second layer of leverage actually is.

This is what makes the second layer structurally different from the first. User-layer leverage scales linearly: each user with AI produces leveraged output, and the value or harm compounds across the organization, but each individual is a discrete amplification. Designer-layer leverage scales exponentially. A single designer's choices propagate to every user of that agent. One person's cognitive priors become the cognitive priors of thousands.

And almost no one with formal authority is doing this work. In practice, the conditions of cognitive leverage inside most organizations are being shaped by vendors, by IT teams with narrow mandates, by power users with no formal authority, and by individual employees configuring custom instructions on their own. The most influential of these - vendors - sit outside the organization entirely. Their master prompts and knowledge-base configurations are shaping how thousands of organizations reason simultaneously, with incentives that have nothing to do with any one of those organizations' strategy.

The parallel here to something I have been writing about - The Hidden Org Chart - is not coincidental. Authority is being exercised by people who do not formally hold it, and consequences flow to people who do not own them. The cognitive architecture, much like the decision authority structure, has been accumulating rather than being designed.

Tolerable today

Right now, this is mostly tolerable. AI is still recent, and we are all still learning as we go. The cognitive priors propagated by agents have not yet compounded long enough for the effects to show up in P&L. The gap between users with rigorous cognitive positions and users with passive ones has not yet calcified into measurable organizational performance difference. Most leadership teams do not feel the asymmetry yet, because the system has not run long enough to surface it.

This is the part of the situation that feels safest. It is also the part that is most misleading.

How the migration actually happens

This is where the connection to Where Advantage Migrates becomes precise. Cognitive leverage amplifies whatever decision architecture is already in place. Pre-AI, the gap between well-designed organizational reasoning and poorly-designed reasoning mattered, but at the margins. With AI applying leverage to whatever architecture exists, that same gap now determines outcomes at scale. The mechanism behind the migration of advantage upward is not leverage alone, and it is not decision architecture alone. It is leverage amplifying architecture.

As AI commoditizes execution - as competitors gain access to the same tools, the same models, the same operational capabilities - the differentiator between organizations no longer lives at the execution layer. It moves upward, into the conditions under which decisions get reasoned about. The conditions of cognitive leverage, set today by mostly invisible actors, are precisely what will determine which organizations compound advantage and which fragment over the next eighteen to thirty-six months.

Organizations that manage cognitive leverage at both layers - who is doing the reasoning, and who is conditioning the reasoning - will produce organizational thinking that compounds across decisions, functions, and quarters. Organizations that have not will discover that the gap between their thinking and a competitor's thinking is no longer something a better strategy or a better operating model can close. The cognitive priors they have absorbed - from vendors with different incentives, from employees with no shared framing, from agents that no one designed - will have shaped the institution itself.

Pre-AI, organizations could afford to leave cognitive architecture implicit. Reasoning was distributed across humans whose priors were roughly consistent with the organization's culture, training, and judgment standards, and the implicit layer worked because the variance was small. AI breaks that assumption silently. The variance is no longer small, the priors are no longer consistent, and the time to design them deliberately is now, while the gap is still closeable.

So what — three questions for executives

If cognitive leverage is real, and if it operates at two layers, three questions follow that most leadership teams cannot yet answer with specificity.

Where in our organization is cognitive leverage being applied today? Not just which tools are deployed - that is the surface question. The deeper question is which decisions and forms of reasoning are being amplified, and on whose cognitive capacity that amplification is multiplying. The user-layer leverage map is not the same as the IT deployment map.

Who is currently conditioning that leverage? Vendors, internal teams, individual employees, default configurations - each implies a different incentive structure shaping how the organization reasons. Most organizations have not yet looked at this layer with intention, which means it has been getting set, mostly invisibly, by whichever actors happened to be in position to set it first.

Are those conditions aligned with our strategy, or have we absorbed them by default? This is the question most organizations cannot yet answer. The cognitive priors compounding in the organization right now were almost certainly not chosen deliberately. The work to align them with strategy - to make the architecture of judgment as designed as the architecture of decisions - has not begun in most places.

Each of these is an addressable diagnostic right now. None of them is addressable retroactively, once the cognitive priors have compounded.

A note on what compounds

The thing I keep returning to is that leverage, in any system, is what amplifies asymmetric outcomes. The investors who survive leverage are those who manage it deliberately. The institutions that will compound under cognitive leverage are those who have made it visible, named it as a dimension of strategy, and begun to manage it - before the gap that separates them from their competitors becomes structural.

This is, in the end, what the migration of advantage actually looks like in practice. Not a forward-looking observation, but a current condition, operating below the surface of every organization deploying AI right now.

The CxOs who act now will design the leverage their organization operates with. The CxOs who wait will inherit what others have designed for them.