Why do we promote the best accountant and then suddenly expect them to know how to lead people?

The same logic appears in almost every function. A strong IT specialist becomes Head of IT. A successful salesperson becomes Sales Manager. An experienced engineer takes responsibility for an engineering team.

There is an obvious reason for this. Expertise is visible. Performance can often be measured. If someone has consistently delivered excellent work, giving them more responsibility feels like a natural next step.

The difficulty is that the new job requires something different.

Research suggests that this is more than an anecdotal problem. A study published in the Quarterly Journal of Economics, based on sales workers across 131 US firms, found that strong sales performance increased the likelihood of promotion, even though pre-promotion sales performance was negatively associated with subsequent managerial performance. The characteristics that make someone excellent at doing the work are not necessarily the ones that make them excellent at enabling other people to do it.¹

Organisations have lived with this tension for a long time. AI may make it considerably harder to ignore.

AI changes where managerial value comes from

There is a tempting argument about the future of management: AI will take care of management tasks while humans concentrate on leadership.

Reality will probably be less tidy.

Management includes planning, coordination, resource allocation, monitoring, decision-making and accountability. Leadership is more concerned with providing direction, exercising judgment, developing people and helping a group work effectively together. In practice, the two overlap.

AI is already becoming useful in parts of both. On the management side, it can summarise reports, analyse performance data, identify deviations, compare scenarios and reduce administrative work. It can also support some leadership activities – for example by helping managers prepare for conversations, explore different perspectives or think through possible decisions.

But that does not mean AI automatically creates better management or better leadership. The same technology can also make it easier to monitor people more closely, generate more metrics and centralise decisions. Whether AI gives managers more space to lead or simply enables them to control more depends largely on how organisations choose to use it.²

So the interesting question is not whether AI will replace managers. It is what remains valuable when some of the activities that used to justify managerial attention become much easier to perform.

For many managers, information has traditionally been an important source of authority. They collect it from different parts of the organisation, make sense of it, check what is happening and pass information up and down the hierarchy. Their position gives them visibility that other people may not have.

AI changes some of that. Information can be aggregated faster. Reports can be summarised almost instantly. Employees can analyse data, prepare alternatives or solve problems without always asking the person above them for the first answer.

That does not eliminate the need for management. It does weaken the case for a manager whose main contribution is being the person through whom information and decisions have to pass.

When the expert becomes the bottleneck

Imagine someone who has spent ten years becoming an excellent accountant.

They know the systems, understand the rules and recognise mistakes almost immediately. Eventually, they are promoted to manage ten accountants.

Continuing to use their expertise seems entirely reasonable. They review an important piece of work here, correct an error there and step in when something becomes complicated. People appreciate their experience, and the manager feels responsible for maintaining quality.

Gradually, however, a pattern develops.

Important work is checked by the manager. Difficult cases travel upwards. Employees learn that it is safer to ask before deciding. Because the manager can usually solve the problem quickly, stepping in often appears more efficient than allowing somebody else to struggle with it.

Eventually, ten people are producing work while one person feels responsible for validating much of it.

The expertise that earned the promotion has become a bottleneck.

AI can magnify this problem. If employees can prepare analyses, reports, proposals and other outputs much faster, the volume of work reaching the approval point can increase as well. Production accelerates while the decision structure remains unchanged.

A manager who believes that everything important still needs to pass through their hands will not automatically become more effective because the team has AI. They may simply become a narrower bottleneck in a faster system.

This changes the value of delegation. Giving people responsibility is no longer only a leadership ideal or an employee-development exercise. It becomes an operating requirement.

Someone still has to decide which decisions belong where, how much autonomy people should have, when escalation is useful and when it merely slows the organisation down. Managers need to create conditions in which people can use greater technological capability without sending every uncertain decision upwards.

That requires leadership, but also something more concrete: thoughtful allocation of decision rights.

Expertise still matters

There is a danger of taking the argument too far.

The future manager cannot simply become a pleasant generalist who leaves expertise to the team and concentrates on conversations.

AI may actually make managerial judgment more demanding. Generative AI can produce convincing analyses that are incomplete, misleading or simply wrong. More information does not automatically produce better decisions. Someone still needs enough understanding of the subject to recognise when an answer does not make sense, to challenge assumptions and to understand the consequences of acting on a recommendation.

Domain expertise therefore remains important.

What changes is how it is used.

A manager does not necessarily need to be the strongest technical expert in the room. They need enough expertise to understand the work, ask intelligent questions and maintain credibility. Their distinctive contribution may lie in combining the expertise of several people, handling conflicting priorities, making decisions under uncertainty and taking responsibility when there is no obviously correct answer.

That is a different role from being the person who knows most.

And it suggests that organisations may need to reconsider how they select managers in the first place.

The promotion system is part of the problem

In many organisations, management remains one of the clearest routes to higher pay, status and influence.

That creates a predictable incentive. Strong specialists who want to progress eventually feel pressure to take responsibility for people, even when their real strength and interest lie in deep expertise.

The organisation then faces an unnecessary trade-off. It risks losing an excellent specialist while gaining an uncertain manager.

A credible expert career path changes that equation. Specialists should be able to gain responsibility, influence, recognition and compensation without people management being the only way upwards.

At the same time, organisations can become more deliberate about what they look for in future managers.

Technical performance will still matter, but it should not be treated as a proxy for leadership potential. The ability to develop other people’s judgment, coordinate across expertise, deal with conflict, allocate responsibility and make decisions when information is incomplete may tell us more about someone’s suitability for management than being the strongest individual performer on the team.

AI makes this distinction increasingly relevant because some of the things organisations used to rely on managers for are becoming easier to produce. Analysis, information gathering and drafting can increasingly be supported by AI. What remains harder to replace is sound judgment, accountability, coordination and the ability to help other people perform well.

The temptation to manage more

There is another possible outcome, however.

Managers whose traditional work becomes easier may not automatically spend the freed capacity on better leadership. Some may respond by increasing their involvement in everything else.

AI makes more detailed monitoring possible. More activities can be measured, compared and reported. Dashboards can become more sophisticated, updates more frequent and performance easier to inspect from a distance.

An organisation could therefore use AI to create more autonomy – or considerably more control.

That choice matters.

A manager with access to increasingly detailed information about a team does not need to act on every piece of it. One of the harder managerial skills may become deciding what not to monitor, where not to intervene and when people should be allowed to exercise their own judgment.

Otherwise, organisations may end up with a strange version of AI-enabled work: employees receive powerful tools that increase what they can do independently, while management uses equally powerful tools to supervise them more closely.

Technology has increased capability on both sides, yet the organisation itself has become no more mature.

So what are managers for?

AI is unlikely to remove the need for managers. Organisations still need priorities, resource decisions, coordination and accountability. Teams still need people who can deal with ambiguity, conflicting interests and difficult decisions.

But AI may force organisations to become clearer about why managerial roles exist.

If the manager’s value lies mainly in collecting information, producing reports, checking routine work and acting as an obligatory approval point, parts of that role will increasingly be questioned – and some should be.

If the value lies in creating direction, allocating responsibility intelligently, connecting expertise, developing other people’s judgment and taking accountability for difficult decisions, the case for the role looks very different.

This brings us back to the accountant who became a manager.

The mistake was never promoting an accountant. Deep expertise can be enormously valuable in leadership. The mistake is assuming that excellence in one job is sufficient evidence of readiness for another, and then designing the new role so that the person continues doing much of what made them successful before.

AI did not create that problem.

It may simply remove some of the reasons organisations have been able to avoid solving it.

And perhaps that is one of its more useful effects on management.

References

1. Alan Benson, Danielle Li and Kelly Shue, “Promotions and the Peter Principle”, The Quarterly Journal of Economics, Vol. 134, Issue 4, 2019, pp. 2085-2134. https://doi.org/10.1093/qje/qjz022

2. Susanne Tafvelin, Maria Forsgren and Andreas Stenling, “Redesigning the Leader Role for the Age of Artificial Intelligence: A Work Design Challenge”, The Journal of Applied Behavioral Science, first published online 20 September 2026. https://doi.org/10.1177/00218863261488834