What should never be automated in HR
Some HR decisions should stay with a human regardless of agent capability: termination, disciplinary outcomes, final pay decisions, and anything requiring judgement about a person's character or intent. Automating the analysis behind these is fine. Automating the outcome removes accountability from a decision that needs someone to own it.
As agents get better at HR tasks, the temptation grows to hand over more of the decision, not just the analysis. Some decisions should never make that move.
The line is accountability, not accuracy
An agent can be highly accurate and still be the wrong owner for a decision. The question is not whether it gets the answer right, but who should be answerable if it is wrong.
- Termination decisions
- Disciplinary outcomes
- Final pay and promotion decisions
- Judgements about a person's character, intent, or trustworthiness
What agents can still do here
Agents can gather evidence, summarise history, flag inconsistencies, and draft documentation. All of that shortens the time a human needs to decide well. None of it should replace the decision itself.
An agent that prepares the case is useful. An agent that closes the case has quietly taken accountability nobody assigned to it.
Drawing this line now, before agents are capable enough to make it tempting to blur, keeps the organisation clear on who is responsible when it matters most.
Underlag
- Our assessment
Employment decisions with legal consequence typically require a documented human decision-maker.
Common questions
- Will this list change as AI improves?
- The boundary is about accountability, not capability, so it should hold even as agents get more accurate.
- Can an agent recommend a termination?
- It can summarise evidence and flag patterns, but the decision should be made and owned by a person.