What Happens to Middle Management When AI Takes Over Routine Tasks

Two stories are circulating about the layer between executives and frontline teams. In one, AI makes middle managers redundant. In the other, AI frees them for more meaningful work and nothing much is lost. If you are trying to work out what happens to middle management when AI takes over routine tasks, neither story is detailed enough to plan against, and both skip the part that matters: what the job is actually made of.

Middle managers spend their time on coordination, reporting, approvals, monitoring, answering questions, coaching and translating strategy into priorities. AI can already handle part of that list and is unlikely to handle the rest well on its own. The result is a role that changes shape, a layer that may become thinner, and a set of organisational design decisions that leaders cannot leave to chance. This guide sets out what the evidence shows, where it runs out, and how to approach the middle layer as a design question rather than a cost line.

What happens to middle management when AI takes over routine tasks?

The administrative and coordinating parts of the role shrink, spans of control tend to widen, and what remains is the work that depends on judgement, coaching and translation between strategy and teams. Roles change before they disappear, and how far they change depends more on how an organisation redesigns the work than on what the technology can do.

In practice the change tends to unfold in a sequence. First, tools absorb tasks such as scheduling, status reporting, first-draft documents and answers to routine questions. Second, managers find that the same team needs fewer hours of coordination, which makes wider spans of control look feasible. Third, the role splits into a smaller set of tasks that still need a person and a heavier concentration of difficult conversations, exceptions and trade-offs. Finally, the organisation either redesigns the role on purpose or finds the gaps later, for example when nobody is coaching new starters or when decisions queue up with a manager now responsible for twice as many people.

This sequence is a pattern, not a law. A layer that mainly relays information looks very different from one that mainly makes decisions, and organisations differ in how much of their managers' time is routine to begin with.

Which parts of a middle manager's job does AI take over?

AI takes over work that is repetitive, rule-bound and based on information that is already digital: scheduling, reporting, tracking, first-line questions and first drafts. It is less reliable where the work depends on context, trust or competing priorities.

Direct evidence on task shift comes from a Harvard Business School working paper, Generative AI and the Nature of Work, by Manuel Hoffmann and colleagues. The researchers used a natural experiment around free access to GitHub Copilot and followed thousands of open source maintainers over two years. Those with access shifted their effort towards core coding work and away from project management. In the main estimates, the coding share of their activity rose by about 5 percentage points, while the project management share fell by about 10 percentage points, roughly a quarter relative to the baseline. They also worked with fewer collaborators and were less likely to seek help from other people.

The authors conclude that the results point to a large potential for AI to transform work processes and possibly to flatten organisational hierarchies. They are also clear about the limits. The setting is open source software, the data covers public repositories, and the people studied are developers doing project management alongside their main work, not managers in a corporate hierarchy. The finding is evidence of direction: when a tool lowers the cost of core work and removes the need to ask others for help, coordination work shrinks. How much it shrinks in a given company has to be measured there.

Does AI change how managers coach and support their teams?

It can, because AI tools now offer real-time guidance that used to come from a manager or an experienced colleague, particularly for newer staff. That reduces demand for routine coaching and escalation, and raises the value of coaching that deals with judgement, motivation and context.

A study of 5,172 customer support agents by Erik Brynjolfsson, Danielle Li and Lindsey Raymond, Generative AI at Work, offers a useful view of this. Access to an AI assistant raised productivity by about 15% on average, and by roughly 30% for less experienced and lower-skilled workers, with small effects for highly experienced workers. The authors observe that traditional coaching in this setting consisted of brief weekly sessions in which a manager reviewed a few conversations, and that AI suggestions delivered in the moment may address a limitation of that approach. They also found that customer requests to speak to a manager fell by almost a quarter after the tool was introduced.

Read together, these findings describe a manager's workload being reduced from two directions: fewer questions from people who now get answers elsewhere, and fewer escalations from customers who are more confident in the person they are speaking to. The authors caution that their results come from a single firm and occupation over the medium run, and should not be generalised across every job. The reasonable inference for managers is narrower. The basic, repeatable parts of coaching are particularly exposed to automation, and the parts that remain are harder and more human.

Do organisations need fewer middle managers?

Often fewer, or with wider spans, but the evidence does not show that the layer can be removed without cost. Whether fewer managers is the right outcome depends on what work is left, how complex it is and how many people each manager can realistically support.

Forecasts that large shares of management roles will be eliminated are expectations, not outcomes, and they tend to assume that coordination is the whole job. A simple illustration shows why wider spans are not automatic savings. Suppose a manager spends 40% of the week on coordination and reporting and AI returns half of that time, which frees about a fifth of the week. If the team then doubles in size, the people-related work that remains, such as one-to-one conversations, development, conflict and hiring, roughly doubles with it. The freed time is absorbed and the manager is left with a heavier job and less margin.

The better question is not how many managers an organisation can remove, but how many people one manager can support well once the routine work has gone. That answer varies by function, by the maturity of the team and by how much of the remaining work is judgement-based. It should come from observing actual work, not from a target ratio.

What work stays with middle managers?

Four kinds of work tend to stay: judgement and exceptions, developing people, translating between levels, and accountability for decisions that AI informs.

• Judgement and exceptions. When a process produces an unclear or borderline result, someone has to decide what it means and what happens next. Managers close to the work are often well placed to do this.
• Developing people. Tools can supply answers and drafts. They are less able to read why a capable person has lost confidence, to judge when someone is ready for more responsibility, or to give feedback that is both honest and well timed.
• Translation between levels. Strategy has to be turned into priorities that teams can act on, and ground-level reality has to be turned into information that senior leaders can use. This depends on knowing both audiences.
• Accountability. When an AI-supported process gets something wrong, an identifiable person needs to own the outcome and the fix. Without this, errors become nobody's responsibility.

None of these is new to middle management. What changes is the proportion of the week they occupy, and the expectation that managers perform them well rather than alongside a heavy administrative load.

What risks come with cutting the middle layer too quickly?

The main risks are losing coaching, translation, local judgement and the leadership pipeline, all of which are hard to see on a dashboard until something fails.

Removing a layer is visible and immediate. The losses are slower and less visible. Newer staff lose a source of context and feedback that a tool does not fully replace. Senior leaders lose a channel through which unwelcome information travels upward. Teams lose a person who can resolve disagreements between functions before they escalate. Decisions that used to be made locally move up the hierarchy, where they take longer.

There is also a pipeline problem. Many senior leaders first learned to lead in middle management roles, where they could practise managing people, handling conflict and making trade-offs at moderate stakes. If the layer is thinned sharply, organisations need another route for people to build those skills, or they will find a gap in their senior ranks several years later. A plan for middle management when AI takes over routine tasks should therefore answer a second question alongside the first: where will tomorrow's executives learn to lead?

What happens to the people in these roles?

Many managers will find the job becomes more people-intensive and less administrative, with higher expectations for coaching, judgement and working with AI. Some will find that shift unsettling.

A manager whose standing came from knowing what was happening, or from controlling the flow of information, loses that source of standing when information becomes available to everyone through tools. The role asks for a different basis of credibility: the quality of decisions, the trust of the team, and the ability to help others perform. That is a real change of professional identity, and it is reasonable for people to feel uncertain about it.

Organisations can make the transition easier by being direct about what is changing and why, by involving managers in redesigning their own roles, and by investing in the capabilities the new role depends on. Those include coaching, difficult conversations, decision-making under ambiguity and fluent use of the tools themselves. Silence tends to be read as a signal that the role is under threat.

How should leaders redesign the middle layer instead of shrinking it?

Start with the work, not the headcount. Map what managers actually do, decide what AI should do, redesign roles, spans and decision rights around what remains, and then build the capability to do it. A strategy can be ready for AI while the organisation is not, a gap explored in this piece on why an AI-ready strategy needs an AI-ready organisation.

A practical sequence looks like this:

1. Audit manager time by task. Ask a sample of managers to record how their week is spent. Separate coordination, reporting and approvals from coaching, decisions and stakeholder work.
2. Classify each task. Decide which tasks AI can do reliably, which it can support with a person reviewing the result, and which should stay fully human.
3. Redefine the role. Write down what a manager is accountable for once the first category has gone, and set spans of control from the real workload rather than a target ratio.
4. Clarify decision rights. Decide who decides what when AI produces a recommendation, and where exceptions go.
5. Build capability. Develop coaching, judgement and AI fluency before the new role is in effect, not after.
6. Protect the pipeline. Create routes for future leaders to practise leading, whether through project leadership, rotations or deputy roles.
7. Measure the outcomes. Track decision speed, error rates, engagement and the experience of newer staff, and adjust.

How can middle managers respond while the design is still being worked out?

Move towards the parts of the job that depend on judgement and relationships, and learn to work with the tools rather than around them.

Begin by tracking where your own time goes, and automate the administrative parts you control. Use the time to invest in what tools do not do well: regular one-to-one conversations, honest feedback, and clear explanations of why decisions were made. Make your translation work visible by showing senior leaders what you are hearing from teams, and showing teams how their work connects to strategy. Ask your own manager how your role is expected to change, and offer a view on which tasks should move first. Managers who take part in the redesign tend to be better placed than those who wait to have it applied to them.

Where should a leadership team start?

Start with an honest audit of manager time, one pilot team and a decision about how future leaders will be developed.

The audit gives you facts to replace assumptions about what managers do. A pilot lets you test a redefined role, a different span of control and new decision rights on a small scale before applying them across the organisation. The pipeline decision keeps the short-term gains from becoming a long-term shortage.

Headlines about middle management when AI takes over routine tasks tend to ask whether the layer survives. A more useful question is what the layer is for, and how the organisation will make sure that work still gets done well. Leaders who answer that deliberately are likely to find the role changes shape without losing its purpose.