The marketing push to frame AI agents as digital employees rather than software tools is producing a measurable and dangerous effect: people become less careful when they think an AI is a coworker, according to new research published in Harvard Business Review.
Emma Wiles of Boston University conducted a study in which managers evaluated work supposedly produced by an AI tool. When the tool was introduced as a “coworker” named Alex with a title and responsibilities, participants found 18 percent fewer errors compared with the same output attributed to a chatbot. They were also 44 percent more likely to escalate questionable work to a manager for further review rather than correcting it themselves, and reported feeling less personally responsible for the output.
The finding has immediate practical implications. Since April 2026, Microsoft, OpenAI, Anthropic and Google have all released tools for managing teams of AI agents, often marketed as digital colleagues. A survey of 1,261 managers found that nearly one-third said their companies already frame AI agents as employees, and 23 percent list them on organizational charts.
Nvidia CEO Jensen Huang has talked about workplaces populated by “digital humans.” The terminology is spreading faster than the evidence that it is useful or safe.
MIT economist and Nobel laureate Daron Acemoglu, who has studied the economics of AI and labor extensively, argues the framing is counterproductive. “AI agents right now are being marketed as things that can replace humans, and I think that’s just a losing proposition,” Acemoglu said. “They should instead be optimized so that they can improve human capabilities, which is not what they have been at the moment.”
The accountability inversion extends beyond productivity concerns. In contexts where errors have real consequences — healthcare, warfare, education, government — treating AI agents as employees creates a scapegoat mechanism. The MIT Technology Review article cites an example from The Guardian: a school bombing in Iran was initially blamed on an AI agent called Claude, when the actual cause was a cascade of human decisions.
Stanford’s Salt Lab took a different approach, presenting 1,500 workers across 104 occupations with information on AI capabilities and asking which tasks they would actually find useful. The results showed a significant gap between what technology experts assume workers want and what workers actually want. For example, experts identified verifying customer credit ratings for sales representatives as an ideal AI task, but the workers themselves said they definitely did not want that task automated.
The research suggests that the most effective role for AI in the workplace is augmentation, not replacement. Treating AI agents as coworkers degrades human oversight and accountability, while treating them as tools that improve human capabilities preserves the judgment that makes oversight meaningful.
Sources: AI agents are not your “coworkers” (MIT Technology Review, June 29, 2026)