Leadership and AI: who decides when the machine recommends?

Category: Decision-Making | Leadership | AI

Reading time: 7 minutes

Leadership and AI now raise a question that appears simple but is not: when a machine makes a recommendation, who is actually deciding?

Artificial intelligence already summarises information, identifies patterns, suggests priorities, assesses scenarios and recommends actions. In many organisations, these recommendations enter management meetings, recruitment processes, performance analyses, commercial decisions and strategic planning.

The problem is not receiving a recommendation from AI. The problem begins when no one can explain what was questioned, what information was excluded and who assumed responsibility for the final decision.

Using AI does not eliminate the need for leadership. It makes leadership more demanding.

A recommendation is not a decision

An algorithmic recommendation may be useful, plausible and well supported by the available data. But, by default, it does not understand the political, human, operational or ethical context in which an organisation operates.

It does not know which relationship will be affected by a change to a team. It does not understand the history of trust between two departments. It does not know the actual implementation capacity of a middle manager. It does not bear the consequences of a failed decision.

This is why the most important distinction is not between human and artificial decision-making. It is between:

• using AI to inform a decision;
• using AI to replace the judgement that the decision requires.

The first can increase capacity. The second can dilute responsibility without anyone noticing.

What research shows about judgement and AI

In a study conducted in three radiology departments in the United States, Lebovitz, Lifshitz-Assaf and Levina analysed how professionals dealt with opaque AI tools in critical decisions.

The researchers identified a relevant pattern. Professionals in only one of the departments consistently incorporated AI results into their final judgement. To do this, they developed questioning practices: ways of connecting their professional knowledge with the recommendation produced by the system.

They did not limit themselves to asking, “what is the answer?”. They sought to understand where the recommendation diverged from their own judgement, which hypothesis might explain the divergence and what additional information was needed before deciding.

The context is medical, not corporate. But the implication for management is clear: an opaque tool does not automatically improve a decision. Quality depends on the practices built around it.

AI can produce a recommendation. But the organisation still needs someone capable of asking: “What is this not seeing?”

Automation and augmentation are not opposing choices

Automation and augmentation are often treated as two incompatible options: either technology replaces a human task or it helps a person perform that task better.

In practice, the relationship is more complex. Automation can free up time and capacity for more demanding decisions. But it can also progressively reduce the team’s exposure to problems that it still needs to know how to interpret.

The risk emerges when speed, cost reduction and output volume become the only criteria of quality. In this scenario, the organisation may use AI to decide faster, but not necessarily to decide better.

Good integration does not only ask, “what can we automate?”. It also asks:

• What human judgement is still required?
• Which decisions require explicit validation?
• Which capabilities must we continue to practise?
• Who is accountable when the recommendation fails?

The resource that emerges between person and machine

A study on the adoption of AI in chess helps make this idea more concrete. By comparing human performance, the performance of AI engines and hybrid human–AI performance, Krakowski and colleagues identified a new decision-making resource created at the intersection between the person and the tool.

The best performance did not depend solely on having the best player or the best system. It depended on the ability to use the recommendation, interpret it and turn it into a decision in the appropriate context.

For organisations, the lesson is direct: the advantage does not lie solely in the tool acquired. It lies in the collective capacity to formulate good questions, interpret recommendations, recognise limitations and decide responsibly.

It is not enough to place AI in the hands of leaders. Organisations must develop the capability to work with it without surrendering human decision-making agency.

Four practices for deciding with AI without delegating judgement

1. Define which decisions may be assisted and which require formal validation

Not all decisions carry the same level of risk. AI can support meeting preparation, the initial analysis of data or the organisation of information. But decisions affecting people, investment, reputation, strategy or legal risk require higher levels of human validation.

The organisation should make clear where AI may suggest, where it may perform tasks and where it can never be the final decision-maker.

2. Institutionalise constructive challenge

Before turning a recommendation into a decision, someone should have explicit responsibility for challenging it.

• What assumptions are embedded in this recommendation?
• What data or perspectives may be missing?
• In which scenario would this suggestion fail?
• Which relevant alternative has not been considered?

The aim is not to delay every decision. It is to ensure that, in relevant decisions, speed does not replace thought.

3. Make the role of AI in the process visible

When an important decision is supported by AI, the team should be able to explain where the tool was used, what information it received, what output it produced and how that output influenced the final choice.

This is not about creating technological bureaucracy. It is about preserving accountability. When no one knows how a recommendation entered the decision, no one can learn from it, correct it or challenge it.

4. Assess the quality of judgement, not only speed

If performance systems value only speed, volume and the immediate achievement of targets, the organisation is encouraging more automatic use of AI.

In leadership, management, HR, operations or strategy roles, the ability to identify risks, challenge assumptions, integrate conflicting information and justify a difficult decision should also count.

This is the difference between measuring results and managing the quality of the process that produces them. As we explore in Organisational performance: why measuring is not the same as managing, indicators are useful, but they only produce improvement when a structured response is built around them.

A checklist before making an AI-assisted decision

• Which specific decision are we trying to support?
• Which part of the process was performed by AI?
• What data, sources or assumptions influenced the recommendation?
• Who questioned it before the final decision?
• Which alternative was considered and rejected?
• Who assumes responsibility if the decision produces an undesirable outcome?

A mature organisation does not use this checklist because it distrusts technology. It uses it to ensure that technology strengthens, rather than weakens, the quality of its decisions.

Leadership and AI: decision-making is becoming more demanding

AI increases the capacity to generate information. It does not eliminate the need to interpret context, manage ambiguity, assume responsibility and choose between competing priorities.

In fact, the easier it becomes to produce answers, the more important it becomes to distinguish a plausible answer from a decision that is appropriate for the context.

This is why leadership development today cannot be limited to communication, feedback, delegation or team management. It must include the ability to work with technology without handing over to technology what remains a human responsibility.

At VAGAR, we help leaders and organisations clarify decisions, develop criteria, strengthen judgement and integrate AI into management practices that continue to place accountable people at the centre.

Explore VAGAR’s work in Organisational Development and Training and in leadership development.

Book an initial conversation

Evidence base

Krakowski, S., Luger, J., & Raisch, S. (2023). Artificial intelligence and the changing sources of competitive advantage. Strategic Management Journal, 44(6), 1425–1452.

Lebovitz, S., Lifshitz-Assaf, H., & Levina, N. (2022). To engage or not to engage with AI for critical judgments: How professionals deal with opacity when using AI for medical diagnosis. Organization Science, 33(1), 126–148.

Raisch, S., & Krakowski, S. (2021). Artificial intelligence and management: The automation-augmentation paradox. Academy of Management Review, 46(1), 192–210.