Thinking with AI: Managerial Judgment
How can managers question assumptions, evidence and alternatives when thinking with AI? A practical approach to stronger judgment and verification.
Ask AI for a strategic recommendation and the response will often be orderly, fluent and persuasive. The managerial work begins next: identifying its assumptions, missing information and alternatives it may have obscured.
Elisa Farri and Gabriele Rosani examine generative AI as a participant in a manager’s thinking process. Their approach emphasizes a questioning dialogue rather than a one-off answer. How AI Can Help Managers Think Through Problems, February 2025.
Stanford’s 2026 AI Index also highlights uneven capabilities across tasks. Strong performance in one area does not establish reliability in another. Managers therefore need evidence relevant to the decision at hand, rather than relying on a model’s general reputation. AI Index 2026, top takeaways and technical performance chapter.
Write down the decision boundary
Before asking for “a good strategy,” prepare a short note: What are we deciding? What do we not know? Which constraints will remain? Who owns the decision? This clarifies the team’s thinking as well as the instructions supplied to the model.
In a market-entry discussion, growth ambitions, distribution capacity and acceptable risk can disappear into the same sentence. Separating them is work to do before asking AI for additional options.
Split the response into three parts
One working practice we suggest at Stratify is to examine every recommendation as a claim, its support and a test:
Part | The manager’s question |
|---|---|
Claim | What exactly is being presented as true? |
Support | Is it based on verifiable evidence or an assumption? |
Test | What new information would make us abandon it? |
Requesting a counterargument can help. But the same system produced that counterargument: two different answers are not two independent sources of evidence. Opening sources, checking company data and involving a relevant expert remain the team’s responsibility.
Record the reasoning alongside the choice
At the end of the meeting, record rejected options, critical assumptions and the trigger for reviewing the decision. This makes it easier to learn both when the outcome is good and when the reasoning proves weak.
The table above is an original practical suggestion, rather than a reproduction of either source. Our Applied AI for Managers program practices this style of questioning through realistic business problems. The aim is to develop the ability to evaluate an answer as carefully as the ability to ask a question.
