AI is rewriting the fundamentals of corporate strategy: algorithms act as cognitive partners, foundation models reorganize ecosystems, and control over critical resources shifts toward stakeholder networks, forcing firms to rethink boundaries, governance and everyday strategic practice.

Main Finding

AI constitutes a theoretical discontinuity for strategic management: it not only changes tools but reshapes the core ontology of strategy. The paper argues that agentic AI and foundation models create hybrid human–algorithmic cognitive architectures, embed algorithmic actors into the microfoundations of firm outcomes, reconfigure ecosystems around shared models and data flows, redistribute resource creation and control across stakeholder networks, and transform strategizing into continuous, AI-augmented practice. Existing strategy theories remain useful but must be reworked into a sociotechnical, multilevel framework that foregrounds learning dynamics, cross-boundary data flows, governance, and ethics.

Key Points

• AI as agentic, general‑purpose technology:
  • Machine learning systems can recommend, act, and adapt; this agentic dimension changes who or what can be a “strategist.”
  • Foundation models (FMs) transcend firm boundaries and create interdependent algorithmic ecosystems.
• Behavioral Strategy:
  • AI augments and alters managerial cognition: attention, framing, heuristics, biases, emotion, and social influence are mediated by algorithmic interfaces.
  • Hybrid cognition emerges (human + algorithmic), changing information processing and decision heuristics.
• Microfoundations:
  • Traditional microfoundations (individual heterogeneity, interactions, emergence) must be extended to sociotechnical microfoundations where AI systems are active agents in interaction networks.
  • Capabilities can emerge via human–AI complementarity and through routines that combine human judgment and algorithmic automation.
• Ecosystems & Platforms:
  • Competitive dynamics shift from firm-vs-firm to ecosystem-of-models and data-flows; platform and data network effects intensify winner-take-most tendencies.
  • Governance, access control, and interoperability of models/data become central strategic levers.
• Stakeholder Resource‑Based View (RBV):
  • Stakeholders (users, data providers, partners, regulators) become co-creators of resources (data, labels, feedback), altering resource ownership and value appropriation.
  • Ethical, political, and distributional issues (privacy, fairness, surveillance) are core determinants of advantage and legitimacy.
• Strategy‑as‑Practice (SAP):
  • Strategy work becomes continuous, embedded in toolchains and routines that integrate AI into sensing, planning, and execution.
  • Material practices, interfaces, and algorithmic governance shape how strategy is enacted.
• Research agenda and normative concerns:
  • Calls for multilevel empirical work, algorithmic governance studies, attention to ethics and inequality, and new measures of AI-capital (models, fine‑tuning routines, data ecosystems).

Data & Methods

• Nature of the paper: conceptual/theoretical synthesis (no primary empirical data).
• Methods used:
  • Integrative literature review across five strategic traditions (behavioral strategy, microfoundations, ecosystems/platforms, stakeholder RBV, SAP).
  • Application of sociotechnical lenses (actor-network theory, sociomateriality, institutional theory) to reinterpret theory.
  • Comparative framing (pre‑AI vs post‑AI perspectives) and proposal of an empirical research agenda.
• Suggested empirical approaches (from the paper’s agenda and implications):
  • Multi-level analyses (individual, organizational, ecosystem, institutional).
  • Mixed methods: qualitative studies of practice, field experiments, archival analyses, event studies, and computational simulations.
  • Focus on measuring new assets (models, proprietary datasets, fine‑tuning pipelines), governance structures, and algorithmic decision flows.

Implications for AI Economics

• Rethinking firm boundaries and capital:
  • Models and curated data pipelines function as a new class of intangible capital; production functions and cost structures should incorporate algorithmic capital and learning returns.
  • Economies of scale and scope in AI (data + compute + model reuse) intensify concentration—requiring new treatments in industrial organization models.
• Market structure & competition:
  • Data-network and model-network effects can create robust incumbency; platform economics must account for shared foundation models and cross-ecosystem complementarities.
  • Antitrust and regulatory analysis should focus not only on prices but on control of data/model access, interoperability, and governance power.
• Value creation and appropriation:
  • Stakeholders supply data and feedback that are value-producing inputs; property rights, bargaining power, and institutional arrangements determine surplus division.
  • Empirical work should quantify how value streams split across model providers, platform owners, complementors, and data contributors.
• Labor and skill complementarities:
  • AI creates nuanced substitution–complementarity patterns: some tasks are automated, others are augmented; microfoundations of task allocation need empirical modeling (e.g., task-level production functions, human–AI complementarity parameters).
  • Policy and welfare analyses must consider distributional impacts of algorithmic augmentation and potential displacement.
• Measurement and empirical strategies for economists:
  • New variables: model quality/architecture, fine‑tuning intensity, proprietary dataset size/uniqueness, API access constraints, latency of model updates.
  • Identification strategies: use of natural experiments (model releases), diff-in-diff on firm adoption, instrumental variables for instrumenting access to models/data, structural dynamic oligopoly models, agent-based and multi-agent RL simulations to study endogenous interactions.
• Governance, regulation, and public goods:
  • Foundation models have public-goods properties and externalities (spillovers, safety risks); economic policy should weigh R&D subsidies, data portability rules, and obligations for model safety/transparency.
  • Pricing and taxation of algorithmic rents (supernormal returns from data/model control) are potential policy levers.
• Research directions for AI economics:
  • Formal models linking algorithmic learning dynamics to market concentration and firm dynamics.
  • Empirical work estimating returns to algorithmic capital and its depreciation/fine‑tuning dynamics.
  • Studying how governance institutions (standards, interoperability, open models) affect competition and welfare.
  • Welfare and distributional analysis of stakeholder co-creation arrangements (e.g., data markets, revenue-sharing mechanisms).

Suggested short list of operational research projects inspired by the paper: - Build a firm-level dataset with measures of AI-capital (model usage, proprietary data, compute spending) and estimate production functions including algorithmic capital. - Event-study analysis of firm performance around major foundation-model releases or API policy changes to identify spillovers and competitive impacts. - Agent-based simulations of multi‑actor ecosystems where firms choose openness/interoperability strategies for models to study emergent concentration and welfare. - Case studies and field experiments on governance interventions (data portability, model auditability) to evaluate effects on innovation and market power.

Overall, the paper reframes strategic advantage in the AI era as an emergent, cross‑boundary phenomenon—rooted in learning processes, data/model ecosystems, and sociotechnical governance—calling for AI economics to develop tools and measurements that capture these new forms of capital, externalities, and distributional effects.