AI Won’t Save Bad Leadership: How AI is Testing the Limits of Leadership

By Igor Voronin

AI is removing many of the limits around information and execution, and in doing so, it is exposing the quality of leadership underneath.

There is a comforting assumption behind much of the investment in enterprise AI: give managers better information, faster analysis and more capable tools, and better decisions should follow.

But what if the problem was never the lack of information?

AI is removing many of the constraints leaders once faced, from slow reporting to costly analysis.  Yet better-equipped managers do not automatically become better leaders.

Stanford’s 2026 AI Index reports that 88 per cent of surveyed organisations were already using AI in at least one business function in 2025. Access is rapidly becoming ordinary. What happens after everyone has the tools?

Why Can Better Tools Reveal Worse Management?

The first wave of enterprise AI concentrated heavily on adoption. Employees needed access, training and permission to experiment. These were sensible priorities, but adoption tells us very little about whether the organisation itself is improving.

Consider what happens when AI enters a company with three competing priorities and no willingness at the top to choose between them. It can produce excellent analysis for all three. If two departments have spent years arguing over ownership, automating the preparation of their reports will not settle the argument. Where managers measure visible activity rather than useful outcomes, AI can generate more presentations, recommendations and updates than anyone has time to absorb.

The result can look surprisingly productive.

Microsoft’s 2026 Work Trend Index, based on 20,000 AI-using knowledge workers, found that organisational conditions, including culture, manager support and talent practices, accounted for 67 per cent of the measured importance associated with reported AI impact. Individual mindset and behaviour accounted for 32 per cent.

The numbers point away from the tool and towards the environment in which it is being used.

“Are our people using AI?” is therefore becoming a less useful management question. A better one is: what decisions, priorities or outcomes are actually improving because they are?

If decisions still climb the same hierarchy, teams still pull in different directions and nobody is willing to stop work that no longer matters, faster analysis changes surprisingly little. AI may even make the underlying problem easier to see. Once producing information is no longer difficult, leaders have fewer places to hide from choices they have avoided making.

The practical test is not adoption alone. Leaders should look for evidence that AI is shortening decision cycles, clarifying priorities or improving outcomes rather than simply increasing the volume of work produced.

When Answers Are Everywhere, Which Ones Deserve Belief?

Leadership has traditionally carried an information advantage. Senior managers saw more of the organisation, received more analysis and had teams available to investigate problems before decisions were made.

AI is weakening that advantage. Information that once travelled upwards through layers of reporting can increasingly be explored by people much closer to the work. A manager’s value cannot rest on simply knowing more than everyone else.

Research by Ben Weidmann, Yixian Xu and David Deming offers a glimpse of what may replace it. In an experiment involving both human teams and AI agents, effectiveness in leading the AI agents was strongly correlated with a leader’s causal effect on human-team performance. Stronger leaders asked more questions, created more conversational exchange and performed better on measures linked to social intelligence and decision-making.

In other words, better tools did not erase the difference between stronger and weaker leaders.

What makes this more difficult is that AI can be persuasive even when it is wrong. Across five experiments involving 1,403 participants, researchers found that incorrect AI advice could reduce decision quality because people did not always reject it. Explanations of how the system reached its recommendation did not reliably remove that tendency.

This creates a harder test than prompting well. What does a leader do when the answer is polished, numerically supported and wrong? What if it confirms something the executive team already wants to believe?

More analysis does not resolve those questions. In some cases, it makes them harder. Leaders need enough scepticism to challenge an attractive answer and enough judgement to know when uncertainty has been disguised as confidence.

A practical discipline is to require leaders to ask three questions before acting on an AI-supported recommendation: What evidence supports it? What evidence would disprove it? What happens if the recommendation is wrong?

Who Owns the Decision When Nobody Really Made It?

The problem becomes even more complicated when AI moves from advising people to managing parts of the work itself.

An OECD study covering more than 6,000 firms across France, Germany, Italy, Spain, Japan and the United States found that algorithmic management tools were already commonly used. Managers often reported benefits to decision quality and their own job satisfaction, but the research also identified concerns around unclear accountability, limited explainability and trust.

Those concerns will become more consequential as AI begins allocating work, evaluating performance, screening applicants, identifying fraud, setting priorities and initiating actions.

Suppose the system follows its instructions perfectly and the outcome is still unacceptable. Who made the mistake?

That cannot be answered by adding a generic “human in the loop”. Leaders have to decide what that human is expected to do, when intervention is required and where responsibility ultimately sits. Otherwise, automation can distribute the decision so widely that everyone participates while nobody quite owns it.

AI is exposing a basic truth about leadership. Much of leadership was never about having the most information in the room. It was about choosing a direction when several were possible, deciding what not to pursue, recognising when the evidence was insufficient and accepting responsibility after the decision was made.

Those tasks become easier to overlook when technology produces answers so readily. A practical governance rule is therefore simple: before automating a decision, leaders should define who owns the outcome, when human intervention is required and what would justify overriding the system.

The next generation of companies will have extraordinary analytical and operational capacity available to them. That does not guarantee extraordinary management.

It leaves leaders with a more uncomfortable question: Now that we can do more, do we actually know what is worth doing?

What Should Leaders Do Differently?

The leadership challenge is not to slow AI adoption, but to ensure that greater capability produces better judgement rather than simply more activity. Three disciplines matter:

• Measure outcomes, not adoption. Track whether AI is shortening decision cycles, improving customer or operational outcomes, or clarifying priorities.
• Challenge confident answers. Require leaders to test the evidence behind AI-supported recommendations and identify what would prove them wrong.
• Keep ownership explicit. Before automating a decision, define who is accountable, when human intervention is required and when the system should be overridden.

Conclusion

AI will increasingly make sophisticated analysis and automation available to every serious competitor, reducing the advantage created by access to the technology itself. The more consequential differences will come from how organisations design authority, evaluate evidence and make trade-offs when machines can generate options almost instantly. In that environment, leadership becomes less about producing answers and more about deciding which questions matter, where human judgement remains essential and how responsibility is preserved as more work is delegated to AI.

About the Author

Igor Voronin is a technology strategist and writer focused on the systems that let high-growth organisations scale with clarity and control. His work spans SaaS architecture, operational discipline, AI and compute infrastructure, with specific interest in how engineering choices shape resilience, speed and long-term business performance.