AI agents manage HR processes, but HRIS doesn't see who made the decision

AI-agents are already making real HR decisions — but HRIS systems still don't know who made them and why.

The Invisible Executor

When an agent rejects a resume, approves a vacation request, or closes an employee's appeal — the HRIS records the fact. But who made the decision, within what authority, and based on what logic — the system doesn't know. In the "author" field, there might be the name of the HR manager who launched the agent, or nothing at all. This creates a gap between how the formal organizational structure looks and how decisions are actually made.

What is a Phantom Organizational Structure

"Phantom organizational structure" is the sum of agents that de facto perform management functions but are nowhere reflected as subjects with authority and responsibility:

• Recruitment agent — filters out 90% of candidates before a recruiter even looks
• Approval agent — approves or postpones vacations according to set rules
• Support agent — closes employee appeals without escalating to a human
• Analytics agent — generates reports on which HR decisions are based
• Onboarding agent — independently guides a new employee through the first weeks

Each of them affects the lives of real people. None of them are listed in the organization chart and none appear in the audit trail.

Why This Creates Risks

The problem isn't the agents themselves, but the fact that HRIS systems were built for people and don't account for agents as decision-making subjects. Agent traces are either lost or attributed to an employee who actually made no decisions.

"The system stores the result but loses the author of the decision,

the boundaries of their authority, and the audit trail" — that's the essence of the problem.

This generates at least three categories of risks.

Compliance: if an auditor asks why a specific candidate was rejected, there's no answer. Fairness: if an agent systematically discriminates based on indirect indicators, detecting this without a complete log of decisions is impossible. Accountability: when something goes wrong, it's unclear who is responsible — the agent's developer, the manager who approved the parameters, or the leadership that implemented the tool.

What to Change Now

The authors of the material propose several concrete steps. First — introduce agent identifiers in the HRIS: each agent should have a unique record describing its authority, limitations, and owner. Second — log agent decisions separately from human actions, linked to the model version and parameters that were in effect at the moment of the specific decision. Third — formally delineate authority: define which decisions an agent makes independently and which require human involvement — and fix this not only in prompts but in regulations.

What This Means

Companies that deploy AI agents in HR without restructuring the accounting system are building management on invisible infrastructure. Sooner or later this ends in an audit, a lawsuit, or simply the inability to explain why the system works the way it does. It's time to think about the structure of agent authority now — while the ghosts haven't yet become the majority in your organizational structure.