Can AI replace the manager? The hard part is owning the decision

Rate this article ↓

The manager who mainly collects updates and forwards them up the chain faces an obvious question: could software perform much of that function? A company has good reason to examine it. Layers of reporting cost money and can slow decisions.

But management also includes choosing priorities, resolving conflicts and accepting responsibility when plans fail. An AI system can change how those decisions are made without making the obligations disappear. The central workforce issue is who does the remaining work after a managerial layer is removed.

Automation is already part of management

An OECD working paper published in February 2025 examined more than 6,000 firms across France, Germany, Italy, Japan, Spain and the United States. It defines algorithmic management broadly: software that partly or fully automates tasks traditionally performed by managers. That category may include AI; it is not synonymous with generative AI.

Surveyed managers perceived benefits for decision quality and their own satisfaction, alongside concerns about accountability, understanding the tools' logic and worker health. These are reported perceptions and adoption evidence, not a randomized demonstration that companies can eliminate management without harm. OECD employer survey.

The International Labour Organization's 2025 exposure assessment also cautions against equating technical exposure with job elimination. Its task-based analysis identifies potential changes to work; it does not count actual layoffs caused by AI. ILO–NASK research summary.

The organization chart can hide the new workload

Imagine a company automates schedules, compiles performance reports and removes a supervisor. The remaining director now oversees more people. Routine administration may fall, yet disputes and unusual cases may concentrate on one desk. This example illustrates a possible tradeoff, not a finding about a named employer.

If nobody measures the time spent resolving exceptions, the company may book a payroll saving while transferring costs to employees and customers. Conversely, if the software handles routine work reliably and escalation remains manageable, a flatter structure could be a real improvement.

This is why the number of managers alone is a poor success measure. A useful evaluation would include response times, rework, employee turnover, customer outcomes and the full cost of oversight.

A score is a choice about what matters

A hypothetical productivity system might reward completed tickets. That can make sense, but a difficult problem may require more time than several easy ones. If employees are judged only by volume, the system creates an incentive to avoid difficult cases.

The underlying problem is the objective selected by the employer. Adding AI does not resolve it. Management must decide which outcomes matter, how to handle missing context and when an exception is justified. It must also distinguish deliberate misconduct from a bad metric.

Giving a human final approval is useful only if that person has the time, information and authority to disagree. A supervisor who receives hundreds of unexplained recommendations may provide a signature without meaningful scrutiny.

The strongest argument for replacing layers

Human managers can be inconsistent, political or simply ineffective. Software may standardize routine decisions and make some patterns easier to examine. Workers should not have to defend a bad supervisor to question a bad algorithm.

The reasonable comparison is between actual alternatives. Test whether a proposed system improves decisions against the existing process, including errors by both. Keep a usable appeal route and examine whether particular kinds of work are systematically undervalued. Those requirements make the efficiency claim more credible.

CivitasPoint assessment

Businesses should be free to remove management work that no longer adds value. They should also identify who inherits each decision, which exceptions require review and how a worker can correct false information.

AI could reduce the need for some reporting and coordination roles. It could also increase demand for people who design processes and handle difficult cases. The balance remains uncertain. A company has not solved management merely because it has moved the decisions into software; it has solved something only when the resulting operation works better.