More companies are putting AI agents on the org chart. New research suggests that decision comes with a catch
A growing number of companies have stopped calling their AI systems tools. They give them names, job titles and a spot on the org chart, and according to new research, that shift comes with a cost nobody planned for. The moment managers start thinking of an AI as a coworker rather than software, they start reviewing its work less carefully.
Emma Wiles, a Boston University professor who studies how AI affects workers, wanted to know what happens to managerial oversight once companies start treating an AI agent as a formal employee, so she teamed up with researchers from Boston Consulting Group to find out. Their paper, Putting AI on the Org Chart: Evidence on Delegation and Oversight, published this July, surveyed HR and finance managers and ran a randomized experiment using identical documents, with only the stated author changing from group to group.
How the experiment worked
The sample was large, roughly 1,200 managers, directors and executives, mostly from the US private sector, all in HR or finance roles that involved reviewing job descriptions or budget documents regularly.
Each participant received five documents containing built-in errors and 20 minutes to review them. Only one detail changed across the three groups. One set of managers was told an AI tool they used had produced the work. Another was told an AI employee named ALEX-3, a direct report, had done it. A third was told a human employee named Alex, also a direct report, had written it.
Wiles explains why that label might matter with a comparison.
"If you think of AI as a tool, like a spreadsheet, and there's a mistake you make in it, you think, oh man, I made a mistake," Wiles said. "But if you were to start calling your spreadsheet Steve, then at some point you might be convinced that Steve made a mistake, as opposed to you were working with this tool and you made a mistake."
Why the label changes behavior
Did giving an AI a job title actually change anything? For most managers, barely. But among the roughly 23% whose organizations already list AI agents on their org charts, it changed quite a bit. The AI employee framing reduced their error detection by about 16% compared with the AI tool framing, according to the paper. Only the human employee group got the careful scrutiny you'd expect.
"I think that framing them as an employee sort of changes their mode of operation where they're no longer thinking, 'I'm working with a tool,' they're thinking, 'I'm delegating to an employee,'" she said.
Wiles said that distinction matters because of where managers place personal responsibility. If they draft something themselves using a tool like ChatGPT, they know their name is on it.
"If I've decided to use ChatGPT to write the first draft of an email, I'm probably going to actually read through it before I send it, because it has my name at the bottom and I am clearly the one responsible for it," she said.
An AI employee breaks that link. Managers tend to assume their organization wouldn't have assigned them a bad one, so they skip the same level of scrutiny they'd apply to their own work.
A different kind of buck-passing
This isn't simply a matter of managers trusting AI's work more. The same managers who caught fewer errors under the AI employee framing were also more likely to ask someone else to double check the work, compared with managers who were told an AI tool produced it. So instead of reviewing the work more carefully themselves, they're passing that job on to someone else.
"They're just kind of kicking it down the road," Wiles said. "It's not that they necessarily think it's so good, they're just sort of like, it's already not my job."
This difference between checking and delegating matters. Plenty of companies deploying AI agents at scale across enterprise workflows are already grappling with how much oversight those systems actually get in practice.
The risk compounds, Wiles said, as more companies give AI agents a formal role inside organizational workflows. "I think the risk is that basically everyone says, well, the agent did it, that's not my problem," she said. "And then you could have worse quality work."
What HR can do about it
None of this means companies should avoid AI agents altogether, according to Wiles. Her recommendation is simpler. Make accountability explicit before the ambiguity sets in.
"You want to be extremely clear about where accountability rests with any work being done by an AI agent," Wiles said. "Every agent is owned by a person, and if the agent does something wrong, that person is going to be the one who's held responsible for it."
Deciding who owns an AI agent's work, and who answers for its mistakes, is ultimately an HR decision as much as a technical one, especially as organizations rethink org charts and reporting lines to accommodate AI agents.
McKinsey's 2025 State of AI survey found that 62% of organizations are already experimenting with AI agents, even as most have yet to redesign the workflows and governance structures around them. Capability, in other words, is outpacing accountability. And according to Wiles's research, that gap traces back to one unanswered question.
Who is actually responsible when the AI agent gets it wrong?
