Sources checked on 12 October 2026.
In a hypothetical scene, a company's leadership is preparing for a meeting on expanding into a new market. Before it starts, everyone receives a tidy memo: the size of the opportunity, the competitors, three scenarios and a recommendation for a pilot launch. The strategy team prepared it with the help of AI. The figures can be traced, the language is clear, and the expected objections are listed.
The tempting question becomes: do we adopt the plan? But nobody has yet settled why this opportunity should come before fixing the experience of existing customers.
When analysis is easy to produce, the answer can arrive before anyone has agreed on the question. The option with the best memo may gain an advantage it does not deserve. Leadership starts here, with defining the decision the company needs and naming the alternatives that will be set aside when it chooses one. A good presentation helps the discussion, but it does not on its own decide what the organisation should do.
In an opinion piece published in October 2026, Diomidis Spinellis imagines a future in which intellectual work recedes as an economic necessity and thinking becomes closer to an optional activity. It is a thesis about the future, not a finding that leadership is disappearing. But it raises a practical question for the present: how does a manager's responsibility change when systems can produce a growing share of the analysis that once justified the manager's position? [1]
A leader's value can rest on the ability to direct commitment: setting the goal, understanding the consequences, allocating resources, then reviewing the decision. These show up in what the company actually chooses. If the memo proposes entering a new market, the decision also covers who will leave other work to serve it, which promise will slip, and how long management will give the hypothesis to prove itself.
Write the decision before you ask for the plan
Asking for the "best growth strategy" leaves wide gaps for the system to fill. What does growth mean? More revenue this quarter, keeping customers, or opening a new source of income? Each goal can produce a different recommendation from the same data. It is better for the brief to start with a specific decision, the limits on resources, the time horizon, and the result management would count as success.
In the expansion example, the brief could be: compare a limited sales pilot in the new market with improving service for the current segment, within the team and budget available. State the assumptions that would change the recommendation, and separate what we know from what needs testing. This request lets the system widen the comparison, and lets the leader see where personal judgement enters the decision.
Then the team needs evidence that does not all come from rewording the documents already available. A call with a customer may reveal a barrier at payment, or the operations lead may explain that delivery requires capacity that does not exist. AI can organise and compare this information, but its existence depends on someone going out into the real world and collecting it. There remains a practical difference between a hypothesis that looks reasonable and a fact that has been tested.
Writing down the critical assumption makes the discussion more precise. Instead of defending the whole plan, the team can test the sentence it rests on: will the customer pay for this service? Can it be delivered at the required quality? And if the answer is no, changing direction becomes a response to known evidence, not a surprise that needs a long justification.
The same data can lead to two decisions
Suppose the new market offers a growth opportunity while existing customers are suffering from service delays. If the company's priority is to expand its presence quickly, it may choose the expansion pilot. If the priority is to stabilise service quality, it may postpone it. Analysis can explain the consequences of both options. The weight management gives each consequence expresses its commitment to customers, employees and stakeholders.
A study published at the CHI conference in 2025 adds a reason to examine how outputs are used. The researchers surveyed 319 knowledge workers and collected 936 examples of using AI at work. Higher confidence in the system was associated with less critical-thinking effort, as reported by the participants. They described a shift in effort towards verifying information, integrating answers and overseeing the task. [2]
These are self-reports in a survey study; they do not show that using AI causes a loss of the ability to think, and they do not test managers' performance specifically. The editorial lesson we propose is narrower: create a clear process for challenging the memo's recommendations. Ask someone responsible for execution to identify what could fail, and ask for evidence that could change the recommendation. Using another model may add a perspective, but agreement between two models is not independent verification against reality.
When a memo shows that one option is better under certain weights, the weights should be shown too. That way management can discuss what it wants, instead of treating the ranking of alternatives as a fact that carries no preferences.
Change what happens in the strategy meeting
The analysis can be sent before the meeting, and the meeting time kept for what needs a shared commitment. The decision owner begins by naming the goal, then explains the strongest alternative they did not choose and what evidence might push them towards it later. The delivery team adds the constraints the memo did not capture. The discussion is then about a specific choice that can be understood, instead of reviewing slides line by line.
A decision needs a clear owner, even if many people helped prepare it. The owner's responsibility includes checking that the evidence is enough for the size of the commitment, making the assumptions visible, and setting the review date. This does not mean redoing every analysis personally. It means being able to explain why they accepted it, its limits, and how they will act if contradicting information appears.
So that the review does not become just a date in the calendar, tie it to a signal from the work. In the expansion pilot, the signal might be the team's ability to serve new customers without harming existing commitments. A suitable threshold is set before the start, and it is agreed what happens if it is not met. Choosing the threshold is itself a management decision that needs a justification; do not give it false precision just because the system suggested a number.
Keep a short record containing the decision, its evidence, its critical assumption, its owner and the date to revisit it. The record helps especially when circumstances change or the team changes. A new leader can understand what was known at the time of the choice, without judging the decision by knowledge that was not available then. AI can be used to update this record and highlight what has changed, with human review of what the changes mean.
At the next meeting, ask the plan's owner to explain what would make them change their mind.
The answer shows whether the organisation has a decision that can be reviewed, or a document looking for approval. As AI makes analysis easier to produce, leaders have more room to focus on this question: what will we commit to now, and on what evidence will we turn back?
Sources
[1] Diomidis Spinellis: Thinking will become a hobby. Opinion piece, 8 October 2026. A starting point for discussion of intellectual work, not a study of the disappearance of leadership.
[2] Lee et al.: The Impact of Generative AI on Critical Thinking. CHI 2025. A survey of 319 participants and 936 examples; the results are self-reports and correlations, not causal evidence.
One email when we publish. Research, product decisions, and what teams report back.
