A few years ago, a good first draft could be evidence of considerable human effort. Research had to be assembled. Options had to be generated. An argument had to be structured. Someone had to sit with a blank page and work out what they thought. Artificial intelligence – particularly generative AI – has changed that experience.

A leader can now produce a plausible market analysis, project plan, risk summary or strategy paper in minutes. The resulting document may be fluent, well organised and full of sensible recommendations.

It may also be based on the wrong question, carry hidden assumptions, miss the institutional context and recommend something the organisation has little ability to execute.

The arrival of an impressive answer does not remove the need for leadership. It changes where leadership adds value.

When answers become abundant

For much of organisational life, authority was reinforced by access to information. Senior leaders attended more meetings, received more reports and had access to people who could conduct analysis on their behalf. The ability to gather information, summarise it and turn it into a presentable document was valuable partly because it required time and specialist support.

AI is reducing some of that scarcity. Information can be assembled more quickly. A larger number of options can be generated. Complex material can be summarised for different audiences. People without extensive technical or editorial support can produce work that, at least initially, looks impressive.

This is potentially liberating. It can widen access to expertise and reduce the time spent on routine intellectual labour. It also creates a new difficulty. A weak argument can now arrive in excellent prose. A superficial analysis can be organised into a convincing structure. Confidence of expression can be mistaken for confidence in the evidence.

Leaders will need to become less impressed by the quality of presentation and more attentive to the quality of reasoning. AI is making answers cheaper. It may make good judgement more valuable.

Asking the question before accepting the answer

AI systems are often most impressive once a problem has been framed. Yet organisations regularly devote enormous effort to answering the wrong questions well.

A fall in performance may be presented as a motivation problem when it is actually caused by conflicting priorities. Delayed decisions may be attributed to weak managers when responsibility has never been properly distributed. Resistance to change may be interpreted as a lack of adaptability when employees have legitimate concerns about an untested proposal.

The first leadership capability is therefore not simply knowing how to use a tool. It is being able to determine what problem deserves attention.

That requires curiosity, but not curiosity in the abstract. It requires the discipline to ask:

• Are we looking at a cause or a symptom?
• Whose description of the problem are we accepting?
• What evidence would challenge our current interpretation?
• Is this a local difficulty or a pattern created by the wider organisation?
• What would we need to believe for this recommendation to be sound?

An intelligent system may help explore these questions. It cannot relieve leaders of responsibility for asking them.

Making sense of competing realities

Organisational decisions rarely emerge from one clean dataset.

Leaders encounter competing accounts of what is happening. Financial evidence may point in one direction while customer experience points in another. An intervention may be technically attractive but culturally difficult. What seems straightforward from the centre may look very different from the frontline.

AI can help organise the evidence. It can identify patterns, compare alternatives and surface information that may otherwise remain dispersed.

But evidence still has to be interpreted.

Sense-making is the ability to construct a sufficiently coherent view of reality to guide action, while retaining enough humility to know that the view may be incomplete. That final qualification matters. One of the dangers of highly fluent AI output is that it can make uncertainty less visible. A polished conclusion may conceal weak data, disputed assumptions or several equally plausible interpretations.

Leaders need to know when the evidence is strong enough to act, when an assumption should be tested and when the organisation needs to remain open to more than one explanation.

Judgement is more than analysis

Leadership judgement has always involved more than choosing the analytically strongest option.

A decision may make sense economically but create an unacceptable ethical consequence. It may be strategically desirable but badly timed. It may work in theory but exceed the institution’s current capacity to execute. It may solve an immediate problem while establishing a damaging precedent. Good judgement takes account of context, timing, values, risk, reversibility and consequence.

This becomes particularly important when AI-generated recommendations appear precise. Precision can be useful, but it can also create an illusion that the messy elements of institutional life have somehow disappeared. They have not.

Leaders still have to decide what matters most. They must weigh interests that cannot be reduced to one measure. They must determine how much risk is acceptable and whose interests may be affected. Crucially, they must accept accountability for whatever follows.

An AI system can contribute to a decision. It cannot carry institutional responsibility for it.

Ethical stewardship is ordinary leadership work

Discussions of AI ethics can quickly become highly technical. Regulation, governance frameworks, data architecture and information security all matter. Yet many ethical questions will arise through ordinary management decisions.

Should this information be used simply because it is available? Should this decision be automated? When is human review necessary? What should an employee or customer be told about how a recommendation was made? Who may be disadvantaged by a model that improves average efficiency? Who is answerable when an AI-informed process causes harm?

These are not questions that leaders can delegate entirely to technical specialists. They concern the kind of institution being built and the standards under which it will operate.

Stewardship may be a useful way to understand this responsibility. Leaders are not merely users of technology. They are stewards of the conditions created through its use. They decide which efficiencies should be pursued, which safeguards are necessary and where human dignity, fairness or accountability must place limits on what is technically possible.

Seeing beyond the immediate task

AI makes it possible to alter one part of an organisation very quickly. What it does not guarantee is that the consequences will remain in that part.

An AI-enabled recruitment process may affect workforce composition, development opportunities, succession and culture. An automated performance measure may alter employee behaviour in ways its designers did not anticipate. Automation in one function may quietly shift risk, work or accountability into another.

This is why systems thinking is becoming more important. A leader must be able to see beyond the immediate output to the wider pattern of consequences. That means understanding interdependencies, feedback loops, incentives and unintended effects.

It also requires an appreciation that organisations are not machines whose components can be optimised independently. They are social systems. People interpret decisions, adapt their behaviour and respond to what is rewarded, tolerated or ignored.

A technically efficient intervention can create an institutionally poor outcome.

Adapting without losing coherence

AI technologies will continue to change. So will their regulation, cost, reliability and public acceptance. Leaders cannot wait for perfect clarity before acting. Nor should they respond to every development with another loosely governed experiment.

Adaptive execution requires a balance. Institutions need to test, learn and revise. They also need enough coherence to know what should remain stable. Values, accountability and strategic intent should not be reinvented every time a new tool appears.

The point of experimentation is to reduce uncertainty, not to avoid commitment indefinitely.

Leaders therefore need to design bounded tests, examine the evidence honestly and stop initiatives that are not delivering. They must be willing to change direction without creating an organisation in which nobody knows which priorities will survive the next leadership meeting.

Adaptability is not constant movement. Sometimes it is the discipline to stay with an important course of action while improving how it is executed.

The risk of cognitive dependence

There is also a more personal leadership risk. AI enables leaders to outsource parts of the thinking process that previously required sustained effort. This can be enormously productive. It can also become a form of avoidance.

If every first-principles argument, difficult synthesis or generation of alternatives is handed over immediately, leaders may become faster while allowing their own reasoning to weaken. Judgement develops partly through the work of thinking: struggling with evidence, constructing an argument, recognising contradiction and learning where one’s initial interpretation was wrong. Leaders should use AI to extend that work, not to escape it.

This does not mean preserving inefficient practices out of nostalgia. Few would argue that leaders should refuse spreadsheets in order to protect their mental arithmetic. But using a calculator wisely still requires some understanding of what is being calculated and whether the answer is plausible. The same principle applies here.

The question is not whether leaders should use AI. They will, and they should. The question is whether they use it as a partner in reasoning or as a substitute for having a considered position of their own.

Developing leaders for what comes next

If these are the capabilities the AI era requires, conventional tool training will not be enough. Leaders certainly need to understand what systems can do, where their limitations lie and how to use them safely. But they also need opportunities to work through ambiguity, competing evidence and ethical consequence.

Leadership development should include difficult cases in which the apparently obvious answer is incomplete. Leaders should practise challenging AI-generated recommendations, distinguishing evidence from confident speculation and deciding where human review is essential.

They should examine real organisational choices, conduct bounded workplace experiments and reflect on what the results reveal. If judgement is contextual, it must be developed in context.

AI will make many aspects of managerial work faster. It may make some activities redundant. It will certainly change where leaders add value. But the hardest questions will remain recognisably human.

What deserves attention? What can be trusted? What is fair? What risk should we accept? What kind of institution are we becoming? Who will take responsibility?

The leadership capabilities of the AI era may not be entirely new. What is new is the speed, scale and consequence with which their absence will be exposed.