Memes move capital. Wars unfold under the gaze of real-time data, traded on prediction markets before the smoke clears. Synthetic influencers command audiences larger than established media brands. Looksmaxxing propagates globally within days once algorithms find their archetype. Clawbots harvest attention across markets. Climate instability surfaces in extreme events with little warning.
This is the operating environment. And it’s strange, recursive, and unstable.
Serious actors are adjusting. BlackRock is shifting from position-based investment logic toward scenario-based strategy, prioritizing AI-safe industries and automating stress testing. The Finnish Innovation Fund Sitra is exploring systemic societal challenges with a human–AI foresight method.
Organizations are expanding analytical approaches to match structural volatility. Risk models grow more granular. Foresight teams make sense of proliferating alternative futures. Leaders use LLMs to synthesize vast textual environments into executive summaries. Strategy teams prompt new visions and missions but rarely generate substantively new trajectories.
We increasingly outsource strategic thinking to machines and ask them what we should do. Some of these approaches help us see the world differently and create new value. But models trained on aggregated discourse tend to stabilize prevailing assumptions. They can extend patterns, yet they struggle to introduce new ones.
This article explores how the practice of strategy relates to the profound shifts now unfolding in automation, generative systems, and machine-assisted meaning making. It proceeds in four steps.
First, it examines how humans and machines reason differently when making sense of the world, identifying where their cognitive capacities diverge. Second, it proposes a useful way of understanding strategy as theory formation, or more precisely as a stack of interdependent theories. Third, it explores how reasoning and theory formation increasingly occur through human–AI cognitive assemblages, where human and machine cognition interact. Finally, it considers how such assemblages might actually produce strategy in practice, speculating how new premises about markets, success, and action can emerge through human–machine collaboration.
The Abductive Gap
In the late nineteenth century, Charles S. Peirce distinguished three modes of reasoning: induction, deduction, and abduction.
Strategic work depends on all three. Induction identifies patterns from observations, mapping trends and signals. Deduction derives necessary conclusions from established premises, clarifying implications and testing coherence. Abduction introduces a new explanatory hypothesis, proposing a new premise about how the system operates.
Large language models operate with exceptional strength in induction and growing competence in deduction. They detect statistical regularities across vast corpora, cluster signals across domains, and trace logical consequences at scale.
The boundary appears at abduction.
Teppo Felin and Mathias Holweg’s Theory Is All You Need: AI, Human Cognition, and Causal Reasoning argues that AI’s data-based prediction differs fundamentally from human theory-based causal reasoning. Strategy and decision making depend on forward-looking, theory-driven belief formation. Valuable strategic insight often requires articulating a causal hypothesis that diverges from prevailing consensus. Models trained on aggregated discourse tend to stabilize that consensus.
Tom Zahavy’s LLMs Can’t Jump from Google DeepMind shows that once premises are given, language models can reason from them. But the formulation of new premises, the conceptual leap that redefines the frame, does not arise from probability distributions over past text. The model remains confined to the statistical structure of its training data.
Induction and deduction operate within a frame. Abduction changes the frame. Strategy advances through such leaps when a new premise reorganizes how the system is understood.
Strategy as Theory
If strategy begins with the formulation of new premises, it is fundamentally a problem of theory.
In his now classic essay What Is the Theory of Your Firm?, Todd Zenger argues that enduring value creation rests on a coherent corporate theory: a causal logic that explains how value is created and guides which adjacencies to explore, which assets to assemble, and which moves generate compounding advantage over time.
Strategy, in this view, is the disciplined formation and pursuit of such a theory. It proposes a structured explanation of how a system actually works and allows leaders to reason counterfactually about how interventions might reshape outcomes.
Zenger illustrates this through Walt Disney’s corporate architecture. Disney’s core capability lay in animated storytelling and the creation of enduring characters. Around that core, the company assembled complementary assets: theme parks, licensing, television, music, and publishing. Films generated characters. Characters populated parks and products. Parks and products reinforced films. Each component amplified the others within a tightly integrated system.
The coherence of Disney’s theory clarified which expansions reinforced the value engine and which diluted it. When leadership drifted away from the animation core, performance weakened. When the company recommitted to the underlying causal logic, growth resumed.
Strategy rarely rests on a single theory. I suggest that organizations operate through at least three interdependent theories:
A theory of market offers a causal account of what is actually unfolding beneath surface symptoms. It reframes the opportunity landscape and clarifies which structural dynamics shape behavior.
A theory of success specifies how value will be created within that reframed landscape. It defines the organization’s distinctive logic of advantage: the configuration of meanings, capabilities, and assets that produces differentiation.
A theory of action translates that hypothesis into a portfolio of interventions designed to test, demonstrate, and extend the causal logic under uncertainty.
Together these three theories form the architecture of abductive strategy, a stacked theory formation under complexity.
Cognitive Assemblage
If strategy is theory formation, the decisive question becomes where new premises originate within a cognitive system that now includes machines.
N. Katherine Hayles describes cognition as layered and distributed rather than unified and exclusively conscious in her essay Modes of Cognition: Implications for Large Language Models. She distinguishes three interacting levels: nonconscious cognition, implicit cognition, and conscious cognition.
Nonconscious cognition operates at high bandwidth and below awareness. It processes information, detects patterns, and generates responses without subjective experience. Hayles argues that cognition does not require consciousness. On this level, machine systems qualify as cognitive agents.
Large language models inhabit this domain. Unlike living organisms, they do not sense a physical environment. They operate within conceptual environments composed of human-authored texts and the representations abstracted from them. Within this domain, transformer embeddings function largely as indexical pointers (correlation-based sign relations) rather than symbolic reference in the human sense. The models detect statistical regularities across vast corpora. They cluster, classify, and transform information. They generate context-sensitive outputs through probabilistic continuation. They interpret through correlation networks and respond flexibly without awareness.
Implicit cognition occupies a middle zone in human experience. It includes embodied learning, habit formation, and tacit integration of experience. It bridges pattern recognition and reflection. An experienced leader sensing that an explanation no longer fits before being able to articulate why is operating in this layer.
LLMs resemble implicit cognition structurally but not materially. They encode regularities and display cross-domain fluency, but their fluency is compression of language use rather than experiential sediment. They have no embodied access to the world and no first-hand contact with consequences. This produces a systemic fragility of reference: coherent language can be generated without reliable grounding when real-world knowledge matters, such as navigation, physical quantities, factual citation, or strategy formation. They do not form habits through lived interaction. They can approximate tacit competence, but they cannot accumulate it through embodied participation.
Conscious cognition operates at lower bandwidth but enables reflection, self-representation, and deliberate evaluation. It is here that explanatory frames are formulated intentionally. It is here that premises are selected and commitments made.
LLMs can simulate reflective discourse. They manipulate symbols. They do not possess awareness. They do not experience insight. They do not commit to premises. Even when they model human intentions with increasing accuracy, they do so from outside the human lifeworld, inferred from patterns in language rather than lived participation.
This layered distinction clarifies the architecture of a human–AI cognitive assemblage. The nonconscious layer expands dramatically through machine pattern detection. The implicit layer remains embodied and human. The conscious layer remains the site of abductive commitment.
Three Strategic Leaps
In strategic practice, this operation of the human–AI cognitive assemblage becomes tangible. When strategy is understood as theory formation, it advances through a series of leaps. In each, the cognitive assemblage integrates nonconscious cognition with implicit and conscious cognition.
1. Reframing the Theory of Market
When prevailing narratives no longer account for market behavior, the frame requires revision. The organization must articulate a new causal account of what is unfolding beneath visible or fluctuating shifts. This is the strategic opportunity.
Here, strategy work starts from making sense of what has and is changing. The AI layer might start with anomaly detection, mapping dominant narratives and revealing where market behavior diverges from belief. The machine layer processes vast informational fields. It maps consensus across earnings calls, industry reports, existing cultural analysis, and consumer data. It surfaces deviations between belief and market behavior and exposes where explanatory models are stretched thin.
But beware: the machine might operate with its own logic, skimming through only parts of the materials and filling in the blanks from what it thinks it already knows through its training data, practically hallucinating research and findings into existence.
Using all this information and operating in interactive dialog with the machine, the human implicit layer challenges the results and recognizes structural tensions, while the human conscious layer double-checks the sources and formulates alternative interpretations and novel causal frames explaining how the market now behaves.
To explore the possibilities, the machine can then model market or human reactions in specific circumstances, drawing on vast secondhand representations of the human Umwelt, the structured field of lived perception. In doing so, it allows us to speculate with possibilities, see our assumptions from a distance, and make more informed decisions on how we frame the strategic opportunity.
2. Reconfiguring the Theory of Success
Once the new theory of the market is formulated, value creation must be reorganized accordingly. This is the strategic trajectory for the organization to reach the opportunity.
The AI layer models structural implications of alternative configurations, surfaces complementarities across the organization’s current capabilities, and scans adjacent possibilities from related fields. It can model and analyze various business models, calculate probabilistic outcomes for each in different market conditions, organize consumer sentiment research, and analyze the data.
The human implicit layer ensures the outcomes are trustworthy and filters this expanded perceptual field. It distinguishes structural signals from episodic fluctuation and draws on lived experience, organizational memory and tacit judgment. The conscious layer determines whether a genuine shift in causal structure is emerging or whether the apparent change is noise.
3. Redesigning the Theory of Action
A theory acquires force only through intervention. This is the strategic reality, to proceed toward the strategic trajectory, to seize the opportunity. The organization structures a portfolio of actions designed to test its causal claims.
AI systems gather and analyze multiple directions and variations of potential interventions and actions, proposing possible constellations and cross-effects. The human implicit layer senses the potential by reading the room and feeling where to intensify commitment, where to recalibrate, and where to revise premises. The conscious layer performs the abductive commitment. It selects a premise and chooses to pursue it under uncertainty. Strategic abduction makes visible the underlying generative structure shaping behavior, capital and meaning. The premise renders the operative system legible.
Through experiments, the premise is subjected to disciplined interrogation. The machine layer stress-tests it across domains, tracing consequences, exposing contradictions and modeling systemic interactions. The implicit layer evaluates whether the causal claims resonate with lived reality. The conscious layer refines, revises or reinforces commitment, then designs interventions that bring the premise into contact with the world.
The Abductive Premium
As AI systems expand inductive and deductive capacity, the center of gravity in strategic work shifts. Pattern detection now scales across domains and analytical synthesis has become widely accessible. Entire industries can generate competent analysis on demand. What does not expand at the same rate is the creative introduction of new causal premises capable of reorganizing the field in which those analyses operate. Scarcity therefore migrates upward, from processing to reframing.
Consider what happened in Glasgow. In 2007, sixty-three young men were killed in gang-related violence in a city then known as the European capital of knife crime. Traditional responses—surveillance, longer sentences, more policing—had been exhausted without moving the numbers. When Karyn McCluskey and the Violence Reduction Unit reframed gang violence not as a crime problem but as a public health epidemic, the system reorganized. Crime logics gave way to illness logics: prevention, early detection, and coordinated intervention across multiple actors. Police officers shifted from proving crimes to building relationships with young men in gang environments.
New resources flowed and new professions gained legitimacy. Over the following decade murders fell by fifty percent, weapon possession dropped by eighty-five percent, and gang-related violence declined by seventy-three percent.
This kind of shift illustrates what Peirce described as abduction: the introduction of a new explanatory premise that reorganizes how a situation is understood. Strategy operates through such moves. Organizations develop working theories of the systems they inhabit, about how markets function, how advantage is created, and which interventions can alter outcomes. In the earlier sections of this article these were described as theories of market, theories of success, and theories of action. Strategy advances when those theories change.
Generative AI alters the cognitive conditions under which these theories are formed. Machine systems expand the inductive and deductive layers of reasoning by mapping patterns across vast informational environments and tracing the implications of premises at scale. Human cognition contributes something different: the ability to sense tensions within those patterns, to interpret them within lived contexts, and to introduce hypotheses that reorganize the frame itself.
This interaction can be understood as a cognitive assemblage. Strategic reasoning increasingly unfolds across intertwined human and machine processes. Machine cognition broadens perceptual range and analytical depth. Human cognition evaluates significance, recognizes structural misfit, and commits to new premises under uncertainty. Within this assemblage strategy becomes a distributed activity in which perception, interpretation, and hypothesis formation circulate across different cognitive layers.
The practical implication is straightforward. As analytical capabilities become widely available, competitive advantage shifts toward actors capable of reframing situations rather than merely analyzing them. Organizations that only optimize within inherited frames improve existing systems. Organizations that introduce new premises reorganize those systems.
Large language models extend the reach of strategic analysis, but the leap that changes the frame emerges from the interaction between machine pattern detection and human hypothesis formation. In that sense, the future of strategy is neither purely human nor purely artificial. It belongs to cognitive assemblages capable of generating abductive leaps and acting on them before they stabilize as consensus.
