Don't Put AI on the Org Chart

The CEOs announcing “AI employees” on their org charts believe they’re sending a signal. Modernity. Scale. Seriousness about the technology. New research from Boston Consulting Group and Boston University, published this month, suggests the signal is real—but pointed in the wrong direction.

The team, led by Matthew Kropp at Boston Consulting Group, ran a randomized experiment with 1,261 managers in HR and finance from the U.S., Canada, and the European Union. Each manager reviewed workplace documents seeded with errors. The only variable changed across the three conditions was the byline: the document came from an AI tool, from a human teammate named “Alex,” or from an AI employee named “ALEX-3.”

Among managers whose companies already list AI agents on their org charts—about 23 percent of the sample—the AI employee framing produced measurable harm across four dimensions and zero benefit on the one outcome leadership keeps justifying it with.

Accountability gets diffused

When the document was framed as the work of an AI employee, personal accountability fell by nine percentage points and accountability attributed to the AI rose by eight. One participant whose company lists “Kevin” as an AI employee described what that looks like in practice: when errors occur, the team narrates them as Kevin’s mistake. Another participant put it more bluntly: “The blame isn’t on a person; it’s on the technology.”

Software cannot bear accountability. It has no agency, no career consequences, no skin in the game. Treating it as if it could is not an oversight—it’s a category error that lets the humans around it stop owning the output.

Review gets worse

Compared with the AI tool framing, managers in the AI employee group asked for additional manager review 44 percent more often, while catching 18 percent fewer errors themselves. Some of the missed errors were the kind that should be caught on a single read: a contract described as cost-reducing while the spreadsheet shows costs going up; an entry-level job description that demands more than 10 years of experience.

The mechanism is intuitive. When the byline says “AI tool,” the reviewer treats themselves as the accountable party. When the byline says “AI employee,” they treat themselves as the second pair of eyes—and second pairs of eyes apply less scrutiny than first ones.

Identity and trust degrade

In the broader sample, managers in companies that use the AI employee framing reported 13 percent more uncertainty about their own professional identity, 7 percent higher concern about job security, and 10 percent lower trust in how AI would be used. One participant put it directly: “If you want people to feel like they will lose their job to AI, or can be easily replaced by AI, then put it on the org chart.”

The result is a leadership signaling problem disguised as a labeling problem. When companies push AI without clarifying what their human employees’ roles become, the path of least cognitive resistance is to assume the AI is the replacement. Some roles are substitutable; many are augmentable. A company that fails to distinguish lets every employee assume they’re in the first category.

Adoption does not move

The justification CEOs offer for the framing is that it accelerates adoption. The data does not support that. Participants exposed to the AI employee framing reported the same intent to adopt as those exposed to the AI tool framing. Across organizations, framing AI as an employee or teammate produced no clear difference in adoption versus framing it as a tool.

What does drive adoption, the researchers found, is managerial modeling. As one participant put it: “At the point that I saw it was becoming tied to employee success—when somebody used an LLM, they got featured at a town hall—I started telling everybody on my team, ‘You’ve got to use this as much as you can.’” A separate Boston Consulting Group study found that companies leading in AI maturity are 3.5 times more likely to have managers role-modeling AI use. Beyond modeling, adoption stalls for reasons most rollouts don’t measure.

What the framing decides

The choice between “AI tool” and “AI employee” sounds linguistic. It is not. It is an architectural decision about how the organization assigns oversight, how it sets expectations for review, and how it explains to employees what they’re being asked to do.

A tool is something a person uses, accountable for the output. An employee is someone the organization governs, accountable in their own right. AI is in the first category and not the second. Putting it in the second silently shifts the accountability model the organization runs on.

Marketing teams have already started running into the same question: as agents take on production work, the marketer’s role moves from producer to director. The Boston Consulting Group team is describing the same shift for every other function. Human roles move up the value chain—toward judgment, oversight, escalation handling, and ambiguity—while agents take on execution. The redesign is what makes the model work. Calling the agent a colleague short-circuits the redesign by implying the role hasn’t actually changed.

The hard part

The framework has clean recommendations. The implementations don’t.

Redefining workflows is a multi-quarter program for any function with real volume. Spans of control don’t expand automatically just because output does; if a manager could oversee eight humans, they cannot oversee eight humans plus 12 agents at the same standard of review. Performance management has to reward the quality of oversight, not just the velocity of output—and most performance systems were never built to measure that. Capability building for managers who now run agentic teams is a curriculum that does not exist yet at most companies.

The point underneath all of it is that the most important AI capability is not technical. It is the judgment to know when to rely on the agent, when to overrule it, and how to structure the work so that the agent’s reach matches its actual reliability. That judgment lives in humans. Treating AI as a colleague obscures it; treating AI as software clarifies it.

. . .

The org chart is a tool for assigning human accountability. It works because the people on it can be reasoned with, promoted, fired, and held responsible. A piece of software can be none of those things. Putting it there does not make the company more advanced. It just makes the accountability harder to find.