An AI system can monitor attendance, allocate shifts or recommend personnel measures. Once its output affects a person’s career, however, formal human confirmation is not enough. A recent case shows that the outcome produced by AI can be strongly influenced by the way a manager asks the question. A company therefore has to define not only the system’s authority but also the human responsibility for how it is used.
The idea of an “AI manager” is no longer purely theoretical. Systems can already help allocate shifts, monitor attendance, evaluate operational data or recommend the next personnel step. It is this last area that shows where ordinary automation ends and managerial responsibility begins.
Forbes describes an experiment by Andon Labs in which a Claude-based system helped manage employees. One worker arrived late for 17 of 23 shifts. The AI initially proposed a formal warning, but after additional information was supplied and a human manager asked whether the employee was genuinely a good fit for the job, it subsequently recommended terminating the employment relationship. This was therefore not an autonomous system decision made without human input. On the contrary, the case showed how strongly a person can influence the conclusion through the wording of a prompt and the choice of context supplied to the system.
For management, that is more important than the technological milestone itself. If a system prepares input for a serious personnel decision, the company must be able to distinguish at least three things afterwards: what facts the system received, what instructions it was given and which part of the final judgment belonged to the human.
A formal “human in the decision process” does not necessarily mean genuine control. A manager may simply approve the system’s output, may unintentionally give it one-sided context, or may phrase the question in a way that practically signals the expected result. In such a case, AI does not become an independent decision-maker but an amplifier of its user’s position.
For personnel decisions, therefore, logging only the final verdict is not enough. It also makes sense to retain important inputs, the system’s recommendation before human intervention and the reasons why the person accepted or changed the recommendation.
It is equally important to separate administrative and evaluative tasks. AI can relatively safely flag that a predefined threshold for late arrivals has been exceeded. Far more sensitive is the decision on whether that behavior justifies dismissal. That judgment involves the circumstances of the case, previous communication, consistency in applying the rules, and legal as well as organizational context.
A company should therefore define the boundary of autonomy in advance for every personnel use of AI. The greater the impact of a decision on a person, the less acceptable it is for a manager to transfer their own responsibility to the system merely because an algorithm generated the recommendation.
KEY TERMS
- Automated management support: AI prepares evidence or recommendations for a decision for which a human remains responsible.
- Leading prompt: Wording of a question that signals the expected or preferred outcome in advance.
- Significant personnel decision: A decision with a substantial impact on a person’s employment status, such as dismissal or a sanction.
- Decision audit: A record of facts, instructions, AI output and the human reasoning behind the final action.