AI can change middle management by preparing information, drafting reports and assisting routine coordination. The more consequential change is what managers are expected to do with that preparation: check the evidence, resolve competing priorities, coach people and take responsibility for decisions.

There is no single inevitable outcome for the management layer. The effect depends on the work, the reliability of the systems and how the organisation redesigns responsibilities. Removing a reporting task does not show that the rest of a manager’s role has disappeared.

For a CEO or HR Director, the useful question is which management work should change, and what support people need to do the revised job well.

Separate information preparation from managerial judgement

Consider the work that surrounds a weekly review. Someone gathers updates, reconciles inconsistent descriptions, prepares the summary, identifies overdue items and asks colleagues what needs attention. An appropriately configured system may help with parts of that preparation.

But a list of overdue items does not explain whether the team has conflicting commitments, an unresolved dependency or a customer who changed the brief. Those possibilities require investigation before the manager reallocates work or challenges someone’s performance.

Coaching and negotiation can also benefit from preparation. A manager might use an approved assistant to organise questions or rehearse a difficult conversation with fictional details. The manager still has to listen, respond to new information and own the agreement reached with another person.

I would therefore redesign the sequence of work before changing the reporting line. Specify which preparation the system performs, which findings need verification and which decisions require a named person. Treat those as distinct responsibilities.

What the evidence does and does not tell us

A 2025 OECD employer survey on algorithmic management covered France, Germany, Italy, Japan, Spain and the United States. Managers reported perceived benefits for decision quality alongside concerns about unclear accountability, difficulty following the tools’ logic and worker health protection.

The research concerns software supporting or automating managerial tasks, which may include AI. It records employer perspectives; it does not prove that AI causes better decisions or predict the removal of middle management. Its findings should not be treated as measurements of Dubai’s workforce.

My practical conclusion is to examine both sides of a proposed change. A faster briefing is useful only if the people relying on it understand its limits and have the authority to investigate. A new responsibility is credible only if the manager has time and resources to carry it out.

A proposed weekly routine for an AI-assisted manager

Imagine a service-operations manager overseeing a team that delivers client projects. This is a hypothetical example, not a case study or a standard programme. The organisation has approved a system to prepare summaries from designated project records. It cannot change client commitments or make employment decisions.

The following routine illustrates how preparation and responsibility could be divided. Adapt the sequence to the team’s working week and operational needs.

Monday: test the summary before setting priorities

The system prepares an overview of project status and flags missing updates. The manager checks the source records for the items that could alter a delivery decision, then asks project owners about unresolved gaps.

Suppose a project is labelled “stalled”. The underlying note may show that the team is waiting for a client’s approval. Treating that label as evidence of poor employee performance would create the wrong conversation.

The manager agrees the week’s priorities with the team and records the reasons for any change. The useful output is an understood set of priorities with owners, not a dashboard everyone has merely received. If the source records are too incomplete to support the summary, the first action is to improve the information.

Tuesday: negotiate capacity and make trade-offs visible

An assistant can assemble the workload information the manager is permitted to use and highlight potential conflicts. It should also expose missing assumptions, such as whether an estimate includes review time.

The manager discusses competing requests with the people affected. If sales wants an earlier delivery while operations lacks capacity, a generated schedule does not settle the commercial trade-off. Someone with the right authority must agree what changes and explain the consequences.

Record the commitment, approver and unresolved risk in the existing work system. Do not allow a tentative recommendation to become a customer promise through an unnoticed automated update.

Wednesday: protect time for coaching

Use any reduction in preparation effort to make room for conversations about obstacles, skills and development. Ask the employee for their account of the work before reaching a conclusion from the operational record.

For example, repeated revisions may reflect an unclear brief or a learning need. The appropriate response could be a better specification, supervised practice or a different allocation of work. The manager needs enough context to distinguish those possibilities.

Keep confidential conversations within the organisation’s permitted arrangements. There is no need to record or upload every discussion simply because software can summarise it. The outcome should be an agreed next step and appropriate support, with only the necessary record retained through approved processes.

Thursday: practise handling exceptions

Review an occasion when the summary, recommendation or workflow did not fit the situation. Perhaps a customer complaint was missing from the source system, or an automated priority conflicted with an agreed commitment.

The manager should know how to pause the affected action, reach the responsible system or data owner and keep the team working safely while the issue is investigated. A route that ends with “contact IT” is incomplete if nobody knows who can authorise a temporary workaround.

Record what failed, the immediate correction and the condition for resuming normal operation. Use the review to improve the process rather than expecting individual managers to compensate indefinitely for a recurring system fault.

Friday: review the decisions, not just the activity

Look at work completed, avoidable rework, unresolved risks and whether important decisions received enough attention. Compare these observations with the purpose of the change.

Ask the team whether the new routine made priorities clearer and support easier to obtain. Also examine whether checking the system created work that was absent from the original business case.

If preparation time was recovered, establish where it went. More useful coaching, quicker resolution of a dependency or better service may be valuable outcomes. They are different from a cash saving. AI Should Augment People, Not Just Reduce Headcount considers how to evaluate those benefits without assuming they arise automatically.

Develop delegation and exception-handling skills

Managers need to describe work precisely enough to delegate it. “Keep me informed” leaves the system’s purpose unclear. A better instruction specifies the permitted sources, reporting period, issues to flag and questions that must remain unanswered when evidence is missing.

Review skills matter equally. Managers should be able to trace an important conclusion to its source, notice conflicting information and decide when specialist advice is needed. They do not need to become model developers to ask those questions.

I would make three responsibilities explicit in the role:

  • Evidence: know which information supports a consequential recommendation and what remains uncertain.
  • Decisions: approve within delegated authority, explain trade-offs and escalate decisions beyond it.
  • People: hear challenges, provide support and ensure an automated description can be questioned.

Assess these through realistic situations. Ask a manager to handle an incomplete report, a disputed recommendation or an exception outside the system’s permitted scope. A fluent presentation about AI does not demonstrate those abilities.

Avoid turning support into inappropriate surveillance

Be clear about the difference between monitoring a workflow and judging a person. A count of completed tasks may omit difficulty, quality, collaboration and help given to colleagues. It should not quietly become an overall performance rating.

I would reject an arrangement in which employees cannot understand what information is being used or raise a correction. Explain the purpose of the monitoring, the decisions it may inform, who can access the information and where concerns can be raised. Have the relevant organisational specialists examine privacy and employment implications for the actual setting.

Managers also need permission to disagree with the system. Requiring approval while treating every override as failure encourages mechanical acceptance. Review the reasoning and outcome instead of rewarding agreement with a recommendation for its own sake.

Change the conditions around the manager

Executives must provide reliable source records, clear decision rights, review time and a workable escalation route. They should also decide which previous reporting duties can stop. Adding AI checks while retaining every old reporting demand is a different workload proposition from the one originally proposed.

Do not assume that a faster reporting process justifies enlarging every manager’s team. Test whether coaching, coordination and exception handling can still be performed at the required standard. Some roles may be redesigned or consolidated; that requires evidence about the whole job and the consequences for the people doing it.

These are leadership decisions, not settings a middle manager can resolve alone. An AI leadership offsite can provide a structured setting to agree the responsibilities, resources and unresolved choices.

Start with one team’s existing weekly routine. Identify a preparation task to test, a decision the manager will retain and a measure of whether the new arrangement helps. Review the experience with both the manager and the team before changing expectations more widely.