Can AI do strategy? A dialogue and debate

Paper highlights

This dialogue and debate was published in the Strategy Science special issue on AI and strategy. Its value is that the contributors expose the different standards hiding behind a yes-or-no question. They disagree over whether AI should be judged by outcomes or human-like reasoning, whether problem framing remains distinctively human, and when strong performance can justify delegation.

The question “Can AI do strategy?” hides a disagreement about what counts as strategy. If strategy is search, prediction, and aggregation, AI already performs parts of it. If it requires causal theorizing, novel problem framing, persuasion, and responsibility for consequences, current systems leave much of the work to people.

Positions in the debate

The essays grew from an August 2025 meeting at the ION Management Science Laboratory, moderated by Todd Zenger. The contributors deliberately do not force a consensus:

• Felipe Csaszar argues that AI can do strategy if the field creates verifiable tasks and systems that expand strategic search, representation, and aggregation.
• Teppo Felin argues that strategy is forward-looking causal theorizing that systems trained on historical data cannot originate.
• Jessica Hullman sees value in ideation, synthesis, and simulation, while retaining human ownership of iterative sensemaking.
• Karim Lakhani emphasizes field evidence and a “jagged frontier” on which AI helps some tasks and hurts others.
• Mari Sako places human problem framing before AI-assisted search and aggregation.
• James Evans argues that internal conflict can improve both organizations and AI systems, while polished recommendations may suppress useful disagreement.
• Ronnie Chatterji expects AI to absorb parts of strategic work but remain a complement where accountability, persuasion, and context matter.

What the debate clarifies

The disagreement separates four questions: Is strategy prediction and search or causal theory and framing? Should AI be judged by its reasoning or outcomes? Does human involvement correct error or reintroduce noise? Which responsibilities should remain human even if performance favors delegation?

Evidence on one question cannot settle the others. Strong performance on a bounded simulation demonstrates capability on that task, but does not by itself justify giving an AI authority over consequential organizational decisions. Conversely, failure to originate a causal theory does not erase useful predictive performance. The collection turns one broad argument into positions that research can test.

How to use this debate

Cite this for

• A compact set of scholarly positions on whether AI can do strategy.
• The different standards behind the question: prediction, causal theorizing, problem framing, judgment, persuasion, and accountability.
• A debate format that separates capability evidence from delegation and responsibility.

Useful for teaching

• Letting students compare outcome-based and reasoning-based standards for AI in strategy.
• Showing why evidence from bounded tasks does not settle authority over consequential decisions.
• Turning a yes-or-no prompt into testable claims about tasks, evidence, and governance.

Careful claim

The debate clarifies where scholars disagree and what evidence would matter; it does not produce a consensus answer to whether AI can do strategy in general.

Other links

• Introductory essay to the “Can AI do strategy?” special issue in Strategy Science
• Table of contents of the Strategy Science special issue on “Can AI do strategy?”