AI agents, the software that does not just answer questions but takes action on its own like updating records or booking travel, are showing up on org charts now. Not as a line item under "IT systems," but as something closer to a colleague. A recent industry report found that most executives already view these systems more like a coworker than a piece of software, and a third of managers describe their AI as a teammate outright.
That framing has a cost, and Boston Consulting Group just measured it. Researchers surveyed more than 1,200 managers and gave two groups the exact same flawed document to review, telling one group it came from an "AI tool" and the other it came from an "AI employee." The group told it was an employee caught fewer errors, forwarded the bad work to colleagues far more often, and blamed the technology instead of themselves.
This is not a small effect confined to one company's culture. It lines up with separate academic research showing that once people sense something like a mind behind a system, they shift blame for mistakes onto that system and let the humans involved off the hook. Give software a name and a face, even a cartoonish one, and people will trust it to catch its own mistakes, which it cannot do.
None of this means agents should stay invisible. The useful part of putting an agent on the org chart is assigning clear ownership: who is responsible for its guardrails, who owns the policy it follows, and who gets escalated a hard call it should not make alone. Anthropic's own research on building agent teams found the same thing, an agent needs a defined role, written context, and a human owner, or it will act on assumptions nobody actually approved.
The part worth dropping is the name. Okta's HR chief has argued that naming an agent reduces fear and restores healthy suspicion, the same way you would question a new coworker. The data says the opposite happens, people relax instead of double checking.
There is a difference between naming a product for customers and naming an internal system employees are supposed to grade. Amazon named its shopping assistant Rufus to feel approachable to shoppers, which is marketing aimed outward. Naming your internal HR bot the same way asks your own staff to extend trust it has not earned, and BCG's numbers say they will take that offer.
For any business rolling out agents this year, the lesson is boring on purpose. Write down who owns the agent, what it is allowed to touch, and who checks its work, then call it "Expense Bot" or "HR Assistant" and leave it there. The moment it gets a name and a seat at the table, your own people start grading its homework less carefully, and that is the opposite of what an agent rollout is supposed to buy you.