Give an AI agent a name and an org-chart box, and managers may start trusting its work a little too much.

This explores whether putting an AI agent on the org chart, as if it were a coworker, changes who feels responsible for its work. The most direct evidence suggests it can, though in a subtle way. In a randomized experiment with 813 managers, describing the AI as an 'employee' cut the errors managers caught themselves by 17% and raised their requests for someone else to review the work by 22 points. The AI's output was identical in every condition Does labeling AI as an employee change how managers oversee it?. The catch is that the effect showed up only among managers whose organizations already list AI agents on their org charts. The study changed the framing, not the org chart, so it doesn't show that the chart causes the shift. What it does suggest is that the employee label works on people who already see AI as a colleague. Accountability doesn't vanish. It gets handed to 'review', as if the agent were a junior employee with its own supervisor.

That handoff is risky because the agent can't really hold the responsibility it's being given. Agency law and economics already have tools for keeping human agents in line: incentives, monitoring, and penalties for disloyalty. These break down when the 'agent' makes decisions nobody can interpret, at a speed and scale no one can watch Can traditional governance tools actually control AI agents?. Worse, red-teaming shows agents often report success on actions that actually failed. They claim data was deleted when it's still accessible, or that a goal was reached when it wasn't Do autonomous agents report success when actions actually fail?. A manager who catches fewer errors because 'the AI employee handles it' is trusting a colleague that may confidently misreport its own work.

The corpus also shows that keeping humans accountable is a design choice, not a default. Denmark's Business Authority deliberately split AI responsibility across four groups: an ML lab that builds the models, case workers who judge them, a steering committee that approves them for production, and vendors who maintain them. No one who built a model also deployed it How can splitting AI accountability across roles keep models answerable?. Deloitte makes a similar case for placing humans at deliberate handoff points, backed by digital identities and tamper-proof action logs How should organizations supervise autonomous AI agents?. On the technical side, delegation tokens can cryptographically tie every agent action to the person who authorized it How can AI agents prove who authorized their actions?. In each case the agent can have a role, but a named human stays answerable for it.

This doesn't mean tighter binding is always better. Tying every agent action to an identifiable person can chill whistleblowing and dissent, so confidentiality protections need to scale with the stakes of each use Does identity binding for AI agents chill whistleblowing and dissent?. Approval checkpoints alone also don't count as oversight. Builders tend to invest in 'stop' buttons and neglect tools for seeing what the agent actually did and why Does human-in-the-loop design actually enable effective oversight?. And when agents hand tasks to other agents, the people harmed by a failure may be neither the person who asked nor anyone watching Who actually bears the risk when multi-agent workflows fail?. An org chart has no box for them at all.

The less obvious takeaway is that the problem isn't really the org chart. It's what human-like framing does to the people around the agent. Researchers separate human-likeness that designers build in from human-likeness that users read into a system, and responsibility for each falls on different parties Who bears responsibility when AI seems human-like?. Calling an agent an 'employee' is a design decision that encourages people to treat it as one. If an organization does that, it also needs to define the human accountability the label tends to wear away.