The First Job of AI-Era Leadership is Protecting Judgment 

By Dr. Hemachandran K, Dr. Raul Villamarin Rodriguez, and Dr. Anderson de Souza Sant’Anna 

As generative AI accelerates knowledge work, leaders must protect the human capacity to question, verify and take responsibility for decisions.  

Every organisation adopting generative AI is measuring the obvious things: hours saved, reports produced, tasks automated and workflows accelerated. These are useful metrics, but they miss the leadership question that matters most. Are people becoming better at thinking, or merely better at accepting polished answers? AI can help teams draft, analyse, summarise and communicate at remarkable speed. Yet if employees stop forming their own view before consulting the machine, stop checking its assumptions or lose confidence to challenge it, the organisation gains output while quietly losing judgment. That trade is more dangerous than it looks. 

The Productivity Dashboard Has a Blind Spot 

Most AI adoption programmes celebrate speed. A team produces a client brief in one hour instead of six. A manager turns a spreadsheet into a summary before the meeting starts. A junior analyst generates ten strategy options before lunch. All useful. But output is not the same as understanding. A polished deck does not prove that the person presenting it understands the assumptions behind it. A fluent recommendation is not a decision merely because it arrives with headings, bullet points and a confident tone.  

The danger is subtle. Knowledge workers can shift from being originators of analysis to becoming editors of machine-generated work. At first, this feels efficient. Later, it can become dependency. Research on automation bias has long shown that people may give excessive weight to automated recommendations, especially when the system appears reliable and the user is under pressure.1 Generative AI intensifies this because it does not only provide an answer. It provides a story that sounds plausible.  

The Real Risk Is Automation Bias 

AI often sounds more certain than it deserves to be. A model can produce a smooth explanation even when the recommendation is incomplete, based on weak assumptions or simply wrong. The language is persuasive. The formatting is clean. The answer arrives quickly. That is precisely why people are tempted to trust it.  

Explanations do not automatically solve the problem. One study found that explanations attached to automated recommendations did not necessarily reduce automation bias and could sometimes increase it.2 This is the explanation paradox: a machine that explains itself may look more trustworthy even when it is still wrong. Leaders should therefore teach teams to separate fluency from accuracy, agreement from validation and explanation from evidence. 

The Jagged Frontier Is Where Judgment Matters 

The most useful evidence comes from a field experiment involving consultants at Boston Consulting Group. The researchers described a “jagged technological frontier”: AI performs very well on some tasks, even sophisticated ones, but performs badly on other tasks that appear similarly difficult.3  

Inside the frontier, AI can make capable professionals faster and better. Outside it, AI can make capable professionals confidently wrong. The practical lesson is simple: people cannot assume the tool knows where its own limits are. This is why judgment matters more, not less. The value of leadership is no longer only to encourage AI use. It is to ensure that AI use does not displace responsibility for thinking. 

Do Not Let AI Become the First Thinker in the Room 

The simplest way to protect independent judgment is to change the order of work. Before an employee asks AI for an answer, they should first write down their own view: What is the central issue? What assumptions are being made? What evidence matters? What would they recommend if the AI tool did not exist?  

Call this blind hypothesis generation. The person develops an initial hypothesis before seeing the machine’s answer. Then AI becomes a tool for challenge, comparison, expansion and testing, not an anchor that quietly defines the whole problem. This is not nostalgia for pre-AI work. It is a guardrail for strategic thinking.  

Use AI as a Thinking Partner, Not an Authority 

The best use of AI is not as an invisible substitute for judgment. It is as a thinking partner that makes judgment sharper. Leaders can ask AI to play four useful roles: a mirror that clarifies an argument, a challenger that makes the strongest case against a preferred option, a simulator that rehearses a difficult conversation, and a scout that surfaces adjacent risks or questions.  

These uses are valuable because they keep human beings in the role of decision-maker. The machine expands the field of view. The human decides what deserves belief, what requires verification and what action should follow. The moment AI becomes the authority, the organisation has confused assistance with accountability. 

Create Friction Where the Stakes Are High 

Leaders often speak about removing friction from work. In routine tasks, that makes sense. Nobody needs a committee meeting to approve a meeting summary. But high-stakes decisions need some friction. A little delay can prevent a great deal of regret.  

One practical approach is a reversibility test. For low-risk, reversible decisions, organisations can automate aggressively. If the outcome can be easily corrected, let AI handle more execution. For high-impact, hard-to-reverse decisions, such as acquisitions, layoffs, capital investments, regulatory disclosures, safety decisions or strategic commitments, organisations should require explicit human ownership. No one-click sign-offs. No “the model recommended it.” The question should be direct: if this decision proves wrong, who will explain it to employees, customers, regulators or the board? 

Make Dissent Routine 

AI can create a quiet form of groupthink. If everybody uses the same underlying model, similar prompts and the same available data, teams can converge on the same elegant but narrow answer. Leaders need to make disagreement a normal part of decision quality. 

A simple RED discipline can help. Recruit independent perspectives before major decisions are finalised. Engage through structured challenge, such as pre-mortems and strongest-case objections. Develop cognitive resilience so that challenging an AI-supported answer is seen as rigour, not disruption. This is not about distrusting every machine output. It is about ensuring that teams do not become too impressed by a machine that sounds as if it attended every management meeting.  

Protect the Apprenticeship 

There is one more issue leaders should not ignore. Junior employees still need to learn how to do the work before they learn how to supervise the machine doing the work. If AI handles every first draft, summary, analysis and calculation, organisations may create an apprenticeship gap. The junior employee looks productive. The output looks polished. But when the unusual case arrives, they may not have the mental models needed to understand what has gone wrong.  

Organisations therefore need selected AI-limited practice zones. Not because AI is forbidden, but because foundational skills must be built before they can be safely delegated.  

The Leadership Test 

AI capabilities will become widely available. Most organisations will have access to similar models, tools and productivity gains. The differentiator will be human judgment. The organisations that win will not be those that automate every decision fastest. They will be those that use AI for scale while protecting the ability of people to notice anomalies, challenge consensus, reason through ambiguity and take responsibility when the machine is wrong.

Disclaimer 

The views expressed are those of the authors and do not necessarily reflect the official policy or position of Woxsen University or its partners. 

About the Authors

Dr. Hemachandran K is Vice Dean of the School of Business and Director of the AI Research Centre at Woxsen University, Hyderabad, India. His work focuses on artificial intelligence, management education, digital transformation and the organisational implications of emerging technologies. 

Dr. Raul Villamarin Rodriguez is Vice President of Woxsen University, Hyderabad, India. His academic and institutional work addresses leadership, higher education, innovation, future skills and the strategic use of technology in business and management contexts. 

Dr. Anderson de Souza Sant’Anna is a Professor at the São Paulo School of Business Administration, Fundação Getulio Vargas (FGV EAESP), Brazil, and Ambassador of the Management Education and Development Division of the Academy of Management (MED-AoM). His research addresses executive education and leadership development.  

Notes 

1. Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors, 52(3), 381-410. https://doi.org/10.1177/0018720810376055

2. Vered, M., Livni, T., Howe, P. D. L., Miller, T., & Sonenberg, L. (2023). The effects of explanations on automation bias. Artificial Intelligence, 322, 103952. https://doi.org/10.1016/j.artint.2023.103952

3. Dell’Acqua, F., McFowland, E., Mollick, E. R., Lifshitz-Assaf, H., Kellogg, K. C., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2026). Navigating the jagged technological frontier: Field experimental evidence of the effects of AI on knowledge worker productivity and quality. Organization Science. https://doi.org/10.1287/orsc.2025.21838