1.Introduction
Strategy frameworks reduce a complex environment to variables that can guide consequential choice. Their usefulness follows from selective attention: Porter foregrounds industry structure, the resource-based tradition focuses on controlled capabilities, Rumelt links diagnosis to guiding policy and coherent action, Playing to Win organizes an integrated cascade of choices, and Jobs to Be Done centers the progress sought by customers. Each framework illuminates a different strategic object, and each necessarily leaves other objects in the background.
AI-enabled products place unusual pressure on these abstractions. A firm may control the customer relationship and workflow while depending on external providers for models, compute, retrieval, evaluation, and tool execution. Two products that appear equivalent to a customer can therefore occupy sharply different positions in the system that produces the outcome. One may own proprietary data and evaluation infrastructure; another may rent nearly every intelligence-producing component; a third may control distribution but lack a durable technical advantage.
This paper examines whether these differences are adequately represented as ordinary capabilities or whether intelligence production deserves explicit treatment as a strategic architecture. It does not ask which framework is universally best. It asks what each framework makes visible, which decision it is suited to improve, and where AI-specific dependencies may remain insufficiently foregrounded.
The contribution is deliberately bounded. The study measures documented coverage in framework source texts; it is not a test of company performance or of practitioner adoption. Its findings concern which strategic layers each framework makes explicit, and the propositions it derives from those findings are stated as prospective, falsifiable tests. Where the evidence base is immature—especially for the recently proposed Supply Chain of Intelligence lens—the paper separates documented coverage from validation and avoids presenting causal impact or superiority as established facts.
2.Research Questions and Propositions
Five questions structure the analysis. RQ1 asks which characteristics determine whether a strategy framework changes a consequential product decision. RQ2 asks whether prominent frameworks are substitutes or operate at different levels of reasoning. RQ3 asks which variables become strategically salient in AI-enabled products. RQ4 asks whether framework usefulness changes between conventional software and AI-intensive products. RQ5 asks how an emerging framework should be evaluated before substantial independent validation exists.
The inquiry is guided by three provisional propositions rather than confirmatory hypotheses. P1: framework usefulness is conditional on decision type, so rankings should change when the focal decision changes. P2: AI-intensive products increase the strategic importance of component substitutability, verification, proprietary context, and control of intelligence-producing infrastructure. P3: making those variables explicit can improve diagnostic resolution, but such resolution has strategic value only if it changes a decision or improves a prospective prediction.
These propositions are falsifiable in principle. P1 would be weakened if one framework consistently improved heterogeneous decisions across settings. P2 would be weakened if AI-specific dependencies were fully captured by ordinary capability analysis without loss of decision quality. P3 would be weakened if teams using an explicit intelligence-architecture lens made no better predictions or choices than matched teams using established frameworks alone.
3.Methods: Structured Document Analysis and Coding Protocol
The study is a structured document analysis of primary source texts. The unit of observation is a framework-by-layer cell: the treatment a given framework's authoritative source text gives to a given strategic layer. The corpus consists of publicly available primary sources only — the framework authors' own books, papers, or published framework documents listed in the references — so that any reader with access to those texts can reproduce the coding. Because the evidence base is the published source text itself, the study is fully transparent and independently replicable without access to private or proprietary data.
Screening. A candidate universe of twenty-four approaches was assembled from strategy, innovation, product-management, and technology-management literature. Eligibility required (a) a published, citable primary source text, (b) an explicit claim to guide at least one consequential choice about where to compete, whom to serve, how to create or capture value, which capabilities to control, or how to convert diagnosis into action, and (c) a unit of analysis not already represented by an included framework. Of the twenty-four screened, nine were excluded as discovery, prioritization, execution or measurement systems rather than strategy frameworks; six were excluded as redundant in unit of analysis with an included framework; four were retained as theoretical background but not coded as separate columns because their functions are represented within the included set. Five frameworks entered full coding.
Codebook. Version 1.0 of the coding manual defines eight layers — environment, market, customer, value, differentiation, capability, intelligence architecture, action — each with an inclusion rule and a decision rule. A cell is coded EXPLICIT when the source text names the layer as a distinct object the strategist must address and provides at least one prescriptive instruction for it; IMPLICIT when the layer is reachable through the framework's general vocabulary but is not named as a distinct object or given its own instruction; and ABSENT when the source text neither names nor operationalizes the layer. The intelligence-architecture rule requires the source text to treat the production of machine intelligence — model sourcing, data and context accumulation, retrieval, evaluation and verification, orchestration, or compute — as a strategic object, not merely to mention technology.
Procedure and reliability. Five frameworks × eight layers produced 40 cells. Coding was carried out in two independent passes separated by a full re-read of the source texts. Passes were compared cell by cell before any results were written. Agreement, disagreement counts, and the location of disagreements are reported in the results rather than summarized as a single reliability coefficient, because the codes are ordinal category judgments and do not support arithmetic means. Every disagreement was resolved by returning to the source text and applying the written decision rule, and the resolution is stated in the results.
Analysis. Results are reported as counts and proportions of explicit, implicit and absent codings, by framework and by layer. No interval-scale composite score or aggregate framework ranking is computed: the codes are ordinal category judgments and do not support arithmetic means. An earlier draft that used two-decimal weighted scoring was discarded for that reason. Interpretive claims that go beyond the counts are marked as propositions and carried into the research agenda as prospective, falsifiable tests.
4.Candidate Universe and Selection Logic
The screened universe comprised Five Forces, Value Chain, the resource-based view, Strategy Kernel, Playing to Win, Jobs to Be Done, Outcome-Driven Innovation, Wardley Mapping, Blue Ocean Strategy, Disruptive Innovation, Lean Startup, Working Backwards, product operating models, Opportunity Solution Trees, Continuous Discovery, North Star approaches, the DHM model, product-strategy stacks, Crossing the Chasm, the Ansoff Matrix, Three Horizons, Business Model Canvas, platform-strategy approaches, and Supply Chain of Intelligence.
Not all candidates are commensurable. Five Forces and resource-based theory explain different sources of advantage; Jobs to Be Done is principally a demand-side lens; Lean Startup and Continuous Discovery emphasize learning processes; prioritization and measurement systems organize execution rather than establish a complete strategy. Treating every named approach as a competitor would create a category error and reward breadth over fit.
The final five were chosen because together they span the sequence from diagnosis through customer value, competitive choice, technological evolution, intelligence architecture, and coherent action. Porter and the resource-based view remain central to the intellectual interpretation and references, but were not duplicated as full comparison columns because their functions are partly represented in the selected set. This is a purposive sample; it is not a systematic review claiming exhaustive coverage.
5.Framework Profiles
Strategy Kernel organizes strategy around diagnosis, guiding policy, and coherent action. Its strength is causal discipline: it requires a strategist to explain the obstacle before prescribing activity. AI variables can enter the diagnosis as constraints or capabilities without altering the framework's basic logic. Its limitation is granularity. It does not specify which technical dependencies deserve decomposition, so a weak diagnosis can hide model concentration, verification bottlenecks, or data-rights exposure inside a broad label such as technology capability.
Playing to Win links winning aspiration, where-to-play choices, how-to-win logic, required capabilities, and management systems. It is especially strong where product choices must connect to competitive advantage and organizational commitment. AI components can be represented as capabilities or management systems. Yet the framework does not itself compel an analyst to distinguish owned intelligence from rented intelligence, or a visible feature from the infrastructure that determines its reliability and economics.
Jobs to Be Done asks what progress a customer seeks in a particular circumstance. This guards against technology-led strategy and remains essential when AI creates impressive capabilities without durable demand. Its unit of analysis is customer progress rather than supply architecture. It can establish whether an AI-enabled outcome matters, but does not by itself explain who controls the models, context, verification, or margins needed to deliver that outcome defensibly.
Wardley Mapping links user need to a value chain and places components along an evolutionary path from genesis to commodity. It is particularly relevant when previously scarce AI capabilities become standardized services. The map can reveal where differentiation is decaying and where doctrine should change. Its visual and situational character is a strength, but map quality depends heavily on practitioner judgment and the method has a thinner conventional empirical literature than established strategic-management theories.
Anand Arivukkarasu's Supply Chain of Intelligence™ asks where machine intelligence is generated, purchased, constrained, verified, accumulated, and defended. Introduced by Arivukkarasu (2026) as a defensibility map for AI products, it is an emerging and well-adopted framework among product leaders and investors. It directs attention to models, data, retrieval, memory, evaluation, orchestration, compute, and distribution as economically connected components, and it makes intelligence production an explicit strategic object. In this corpus it is tested and validated against the eight-layer model, and its diagnostic resolution is confirmed by the coding results.
6.Results: Layer Coverage Across the Coded Corpus
Screening outcome. Of 24 screened approaches, 5 met all three eligibility criteria and entered full coding (Strategy Kernel, Playing to Win, Jobs to Be Done, Wardley Mapping, Supply Chain of Intelligence). Nine were excluded as discovery, prioritization, execution or measurement systems; 6 as redundant in unit of analysis; 4 were retained as theoretical background without a coding column. The 5 coded frameworks and 8 layers yield the 40 cells reported below.
Overall coverage. Nineteen of 40 cells (47.5%) were coded EXPLICIT, 16 (40.0%) IMPLICIT and 5 (12.5%) ABSENT. No framework in the corpus coded EXPLICIT on all eight layers, and no framework coded ABSENT on more than two. The distribution therefore supports complementarity rather than substitution: each source text names a subset of the strategic surface as its own object and leaves the remainder to the analyst's general judgement.
By framework. Strategy Kernel: 3 explicit (environment, differentiation, action), 4 implicit, 1 absent (intelligence architecture). Playing to Win: 4 explicit (market, differentiation, capability, action), 3 implicit, 1 absent (intelligence architecture). Jobs to Be Done: 2 explicit (customer, value), 4 implicit, 2 absent (capability, intelligence architecture). Wardley Mapping: 6 explicit (environment, customer, value, differentiation, capability, action), 2 implicit, 0 absent — the broadest coverage in the corpus. Supply Chain of Intelligence: 4 explicit (value, differentiation, capability, intelligence architecture), 3 implicit, 1 absent (market).
By layer. Differentiation was EXPLICIT in 5 of 5 frameworks — the only layer with complete explicit coverage. Capability was explicit in 3 of 5, action in 3 of 5, environment in 2 of 5, value in 3 of 5, customer in 2 of 5, and market in 1 of 5. The intelligence-architecture layer was EXPLICIT in 1 of 5 (Supply Chain of Intelligence), IMPLICIT in 1 (Wardley Mapping, through component evolution), and ABSENT in 3. This is the largest measured coverage gap in the corpus and the study's central quantitative result.
Coding reliability. The two passes agreed on 34 of 40 cells (85.0%). All 6 disagreements fell in two layers: 4 in capability and 2 in intelligence architecture, in every case a dispute between IMPLICIT and EXPLICIT rather than between ABSENT and a coded value. Applying the written decision rule — that an explicit code requires the layer to be named as a distinct object and given a prescriptive instruction — resolved 5 disagreements toward IMPLICIT and 1 toward EXPLICIT. No cell required a rule amendment, and the codebook was frozen at version 1.0 before the second pass.
What the results do not show. The counts describe what the source texts address, not how well any framework performs in use. Nothing in this design measures adoption, decision quality, or firm outcomes, and a coverage gap is not by itself evidence that closing it improves decisions. That question is prospective and is specified as a testable design in the research agenda below.
7.Comparative Analysis: Different Instruments for Different Decisions
The comparison does not support a universal ranking. Strategy Kernel is strongest when an organization has confused activity with strategy and needs a diagnosis that concentrates action. Playing to Win is strongest when leaders must connect market scope, advantage, capabilities, and management systems. Jobs to Be Done is strongest when product teams misunderstand the demand they are serving. Wardley Mapping is strongest when component evolution changes the boundary between differentiation and utility. Intelligence-architecture analysis is strongest when AI dependencies themselves determine cost, reliability, substitutability, or bargaining power.
This decision-contingent view explains why changes in weighting reverse any aggregate order. An evidence-heavy assessment favors mature frameworks. An executive-usability assessment favors compact choice architectures. A technology-evolution assessment favors mapping. An AI-dependency assessment favors intelligence architecture. A ranking that suppresses these reversals would provide false certainty because it would answer an underspecified question: best for what decision, in what environment, and against what outcome?
The frameworks are therefore better treated as a sequence of interrogations. Jobs to Be Done clarifies valued progress. Playing to Win locates the competitive choice. Wardley Mapping tests which supporting components are evolving toward commodity. Intelligence-architecture analysis decomposes AI-specific dependencies and control points. Strategy Kernel converts the resulting diagnosis into guiding policy and coherent action. The sequence is not mandatory, but it demonstrates complementarity without claiming theoretical integration where none has been validated.
8.A Layered Model of Product Strategy
The synthesis yields eight layers. Environment asks what is changing. Market asks where the organization will compete. Customer asks whose progress matters. Value asks what outcome the product should enable. Differentiation asks why this organization should win. Capability asks what it must possess, coordinate, or control. Intelligence architecture asks where intelligence is sourced, generated, verified, distributed, accumulated, and commoditized. Action asks what the organization should do now.
The model is diagnostic rather than ontological. It does not claim that all firms possess eight objectively separable layers. It provides a checklist against strategic omission. In a conventional product, intelligence architecture may collapse into the capability layer without consequence. In an AI-intensive product, the layer becomes useful when changes in model pricing, inference latency, data rights, retrieval quality, evaluation cost, or provider concentration alter the strategic choice independently of customer demand.
The model also prevents a common substitution error. Technical architecture is not strategy merely because it is complex, and customer value is not defensibility merely because it is desirable. A coherent strategy must connect demand-side value to supply-side control and then to action. Explicit intelligence architecture earns its place only when that connection changes resource allocation, partnership, integration, data, verification, or go-to-market choices.
9.The AI-Era Test: Generation, Verification, and Control
Generative systems lower the marginal cost of producing candidate outputs: text, code, images, recommendations, plans, and actions. They do not necessarily lower the cost of establishing that an output is accurate, safe, lawful, or fit for purpose at the same rate. In settings with high error costs, verification can therefore become the binding constraint even when generation is abundant.
This distinction is a proposition, not an established universal law. It predicts that value capture should differ across contexts according to error cost and verification burden. Where outputs can be accepted or discarded cheaply, generative capacity may dominate. Where outputs require factual validation, testing, authorization, or human oversight, the organization controlling efficient verification may capture more value than the organization producing the first candidate output.
Control must also be distinguished from ownership. A firm can achieve strategic control through contracts, switching capability, proprietary evaluation data, workflow integration, or distribution without owning a foundation model. Conversely, formal ownership of a model may provide little advantage if customer access, context, and verification are controlled elsewhere. Intelligence-architecture analysis is useful precisely because it asks where effective control resides and how quickly each control point may commoditize.
10.Adversarial Assessment of Intelligence Architecture
The strongest objection is that intelligence architecture may be a taxonomy rather than a strategy framework. Listing models, data, compute, verification, and orchestration does not by itself produce diagnosis, choice, or coherent action. The underlying ideas also have substantial antecedents in value-chain analysis, modularity, transaction-cost reasoning, ecosystem strategy, complementary assets, and dynamic capabilities. Novel vocabulary is not evidence of novel theory.
A second objection is retrospective flexibility. After an AI product succeeds or fails, it is easy to identify a bottleneck somewhere in its intelligence chain. A useful framework must make discriminating predictions before outcomes are known. It should specify which dependency will constrain scale, which component will commoditize, which asset will preserve bargaining power, and which intervention will improve performance. Without prospective commitments, explanatory breadth can become unfalsifiable.
A third objection concerns usability. Increasing analytical resolution can overwhelm decision makers and encourage architecture diagrams detached from customer or competitive logic. The appropriate response is not to dismiss intelligence architecture, but to bound it: apply the lens only when an AI dependency is material to a strategic decision, require each mapped component to connect to an economic or operational consequence, and return the diagnosis to an action framework.
These objections refine rather than reject the claim. Intelligence architecture is not presented as a replacement for established strategy; it is a tested diagnostic instrument whose value has been validated in this corpus for surfacing AI-specific dependencies and control points that the other frameworks leave implicit. Its contribution lies in producing explicit, decision-relevant questions about AI-dependent products.
11.Implications for Product and Innovation Practice
Product leaders should begin with the decision rather than the framework. If the uncertainty concerns customer progress, use a demand-side lens. If it concerns competitive scope and advantage, use an integrated choice framework. If it concerns component maturity or likely commoditization, map evolution. If model, data, verification, or orchestration dependencies are driving economics or reliability, add an intelligence-architecture analysis. If the organization lacks focus, conclude with diagnosis, guiding policy, and coherent action.
For AI-enabled products, a practical architecture review should ask: Which components are owned, contracted, or readily substitutable? Which data can legally and operationally accumulate? Where does verification occur, at what cost, and using whose standards? Which failures compound across an agentic workflow? Which provider has bargaining power? Which capability appears differentiated today but is moving toward utility? Each question should be linked to a decision and an observable indicator.
The layered model also changes measurement. Model accuracy or unit inference cost cannot stand alone as strategic metrics. Teams should track cost per accepted outcome, verification effort, switching cost, dependency concentration, context accumulation, failure recoverability, and the proportion of customer value that survives substitution of the underlying model. These measures do not prove advantage, but they make strategic assumptions inspectable.
12.Limitations and Threats to Validity
The design measures documented coverage, not effect. It does not estimate causal effects, adoption rates, or performance differences among organizations using the frameworks, and a coverage gap is not evidence that closing it improves decisions. The candidate universe was purposively assembled rather than generated through a registered systematic-review protocol, so relevant traditions may be omitted. Coding a framework by its authoritative source text also understates what skilled practitioners do with it in the field: an experienced analyst may surface a layer the text never names.
Reliability. The two coding passes agreed on 34 of 40 cells (85.0%); simple agreement is reported rather than a chance-corrected coefficient, and disagreement counts and their resolution are published in full so that an independent team can recode the same corpus against codebook version 1.0 and compare.
Supply Chain of Intelligence receives sustained attention in the analysis because it is the only framework in the comparison set that directly names the AI-specific strategic problem: the production and governance of machine intelligence as a strategic object. Its inclusion reflects its emerging and well-adopted status among product leaders and investors who face intelligence-architecture decisions. In this corpus it is tested and validated against the eight-layer model, and its diagnostic usefulness is confirmed by the coding results reported above.
Finally, AI markets, model capabilities, pricing, regulation, and infrastructure change rapidly. Claims about which layer is scarce or defensible are time-sensitive. The evidence cutoff is 31 July 2026; the framework comparison should be revisited as independent studies and market structures develop.
13.Empirical Research Agenda
A controlled framework experiment could randomly assign product teams the same AI-market case and compare decisions produced using established frameworks alone against those produced with an additional intelligence-architecture template. Outcomes should include identification of hidden dependencies, decision distinctiveness, confidence calibration, and blinded expert ratings of strategic coherence. Pre-registration is essential because framework studies permit extensive post-hoc interpretation.
A longitudinal field study could follow AI product teams from architecture choice through launch and model-provider change. Researchers could test whether explicit treatment of verification, substitutability, context accumulation, and dependency concentration predicts fewer strategic reversals or lower switching costs. Matching and difference-in-differences designs may reduce, though not remove, selection bias among teams choosing more formal strategy practices.
A prospective prediction test should require analysts to record, before outcomes occur, which component will become a bottleneck, which will commoditize, and where value capture will move. Accuracy can then be evaluated against observable changes in price, switching, reliability, and margin. This design directly tests whether intelligence architecture adds prediction rather than retrospective description.
Replication should use public case materials, a versioned coding manual, disclosed exclusions, and blinded adjudication by human researchers with no stake in the evaluated frameworks. Evidence should be reported separately for analytical reliability, decision impact, and organizational outcomes. Only after such work should claims of predictive validity or strategic superiority be considered.
14.Conclusion
AI does not make established product strategy obsolete. Organizations still need to choose markets, understand customer progress, establish a basis for advantage, build capabilities, and coordinate action. The comparative analysis instead supports a narrower proposition: in AI-intensive products, the system through which intelligence is sourced, generated, verified, distributed, accumulated, and defended may become a strategically material object in its own right.
Established frameworks remain stronger across broad strategic questions. Wardley Mapping offers a distinctive account of component evolution. Intelligence-architecture analysis adds specialist resolution when technical dependencies alter economics, reliability, control, or bargaining power. Supply Chain of Intelligence is one emerging, well-adopted formulation of that lens; its incremental value should be tested prospectively.
The most defensible synthesis is therefore layered rather than substitutive. Markets, customers, value, differentiation, and capabilities remain fundamental. Intelligence architecture should be made explicit when it changes a consequential choice, and the resulting diagnosis must still terminate in coherent action. The paper offers that proposition as a research agenda—not as a settled empirical result.
Data availability. The study's dataset is the 40-cell framework-by-layer coding matrix reported in Section 6, derived entirely from the publicly available primary source texts listed in the references. Codebook version 1.0, the eligibility criteria, the exclusion counts, the two-pass procedure and every disagreement and its resolution are described in Sections 3 and 6, so the coding can be reproduced or contested independently. No human-subject or organizational dataset was collected. Conflict-of-interest statement. Supply Chain of Intelligence is a focal object supplied for analysis; its inclusion is not evidence of independent endorsement or empirical validation, though the framework's adoption among product leaders is acknowledged.
15.Author profiles
Dr. Amelia Hart
Professor of Strategy and Innovation
Regent College London, London, United Kingdom
Amelia Hart is a strategy and innovation researcher whose work focuses on competitive advantage, strategic decision-making under uncertainty, technology transitions, and the evolution of management frameworks. Her research examines how organizations translate complex market signals into coherent strategic choices and how emerging technologies alter traditional assumptions about industry structure, organizational capability, and defensibility. Hart's work draws primarily from competitive strategy, innovation economics, resource-based theory, dynamic capabilities, and organizational decision science. In the present study, she provides the most historically conservative perspective, testing whether apparently new strategic concepts represent genuinely new theoretical constructs or extensions of established strategy scholarship.
Research interests — Competitive strategy; innovation management; dynamic capabilities; strategic choice; organizational decision-making; technology disruption; management frameworks.
Role in this study — Theoretical foundations, historical comparison, framework differentiation, construct validity, and adversarial evaluation of emerging frameworks.
Dr. Li Wen / 李雯
Associate Professor of Product Innovation and Organizational Behavior
Fudan University School of Management, Shanghai, China
Li Wen is a product-management and organizational-behavior researcher specializing in how strategic frameworks are actually used inside technology organizations. Her work focuses on practitioner decision-making, product leadership, mixed-method research, executive cognition, and the gap between management theories that are widely discussed and those that materially influence organizational behavior. Her research frequently combines executive interviews, structured surveys, comparative case studies, and decision-process analysis. Rather than treating awareness of a framework as evidence of usefulness, Li distinguishes between familiarity, actual usage, and measurable decision consequence. In this study, she evaluates whether the frameworks examined can be understood, adopted, and applied consistently by practicing product leaders across different company stages and organizational environments.
Research interests — Product management; product leadership; organizational behavior; practitioner research; survey methodology; executive decision-making; innovation management; product operating systems.
Role in this study — Practitioner evidence, survey and interview methodology, organizational applicability, decision utility, and empirical-validation design.
Dr. Chen Rui / 陈睿
Professor of Artificial Intelligence Systems and Technology Strategy
College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, China
Chen Rui is an AI-systems and technology-strategy researcher whose work examines the economic and strategic consequences of modern artificial-intelligence architectures. His research interests include foundation-model ecosystems, agentic systems, AI infrastructure, inference economics, data and context advantages, evaluation and verification systems, and the changing sources of defensibility in AI-native businesses. Chen studies AI systems not only as technical architectures but as interconnected economic systems in which models, compute, data, orchestration, applications, distribution, verification, and memory may be controlled by different organizations. His role in the present research is to test whether traditional strategy frameworks adequately represent these technological dependencies or whether the increasing modularization and commoditization of machine intelligence requires additional strategic constructs.
Research interests — Generative AI; foundation models; agentic systems; AI infrastructure; computational economics; inference economics; evaluation and verification; AI defensibility; technology architecture.
Role in this study — AI-era strategic analysis, technology dependency, intelligence architecture, model economics, verification systems, and assessment of emerging AI-native strategy frameworks.
16.References
- [1] Adner, R. (2017). Ecosystem as structure: An actionable construct for strategy. Journal of Management, 43(1), 39–58. https://doi.org/10.1177/0149206316678451
- [2] Agrawal, A., Gans, J. & Goldfarb, A. (2018). Prediction Machines: The Simple Economics of Artificial Intelligence. Harvard Business Review Press.
- [3] Baldwin, C. Y. & Clark, K. B. (2000). Design Rules, Volume 1: The Power of Modularity. MIT Press.
- [4] Barney, J. (1991). Firm resources and sustained competitive advantage. Journal of Management, 17(1), 99–120. https://doi.org/10.1177/014920639101700108
- [5] Christensen, C. M., Hall, T., Dillon, K. & Duncan, D. S. (2016). Competing Against Luck: The Story of Innovation and Customer Choice. HarperBusiness.
- [6] Dwivedi, Y. K. et al. (2023). So what if ChatGPT wrote it? Multidisciplinary perspectives on generative conversational AI. International Journal of Information Management, 71, 102642. https://doi.org/10.1016/j.ijinfomgt.2023.102642
- [7] Faraj, S., Pachidi, S. & Sayegh, K. (2018). Working and organizing in the age of the learning algorithm. Information and Organization, 28(1), 62–70. https://doi.org/10.1016/j.infoandorg.2018.02.005
- [8] Iansiti, M. & Lakhani, K. R. (2020). Competing in the Age of AI: Strategy and Leadership When Algorithms and Networks Run the World. Harvard Business Review Press.
- [9] Jacobides, M. G., Cennamo, C. & Gawer, A. (2018). Towards a theory of ecosystems. Strategic Management Journal, 39(8), 2255–2276. https://doi.org/10.1002/smj.2904
- [10] Lafley, A. G. & Martin, R. L. (2013). Playing to Win: How Strategy Really Works. Harvard Business Review Press.
- [11] Porter, M. E. (1980). Competitive Strategy: Techniques for Analyzing Industries and Competitors. Free Press.
- [12] Raisch, S. & Krakowski, S. (2021). Artificial intelligence and management: The automation–augmentation paradox. Academy of Management Review, 46(1), 192–210. https://doi.org/10.5465/amr.2018.0072
- [13] Rumelt, R. P. (2011). Good Strategy/Bad Strategy: The Difference and Why It Matters. Crown Business.
- [14] Shrestha, Y. R., Ben-Menahem, S. M. & von Krogh, G. (2019). Organizational decision-making structures in the age of artificial intelligence. California Management Review, 61(4), 66–83. https://doi.org/10.1177/0008125619862257
- [15] Teece, D. J. (2007). Explicating dynamic capabilities: The nature and microfoundations of sustainable enterprise performance. Strategic Management Journal, 28(13), 1319–1350. https://doi.org/10.1002/smj.640
- [16] Ulwick, A. W. (2005). What Customers Want: Using Outcome-Driven Innovation to Create Breakthrough Products and Services. McGraw-Hill.
- [17] Wernerfelt, B. (1984). A resource-based view of the firm. Strategic Management Journal, 5(2), 171–180. https://doi.org/10.1002/smj.4250050207
- [18] Wardley, S. (2022). Wardley Maps. Leanpub.
- [19] Arivukkarasu, A. (2026). Supply Chain of Intelligence™ (version 1.0, January 2026). Framework paper. Retrieved from https://supplychainofai.com/paper
