Emma Wiles, a business professor at Boston University, found that managers caught 18% fewer errors when they treated an AI tool as a chatbot rather than an agentic employee. This discovery challenges the growing trend of Silicon Valley tech giants, including Microsoft, OpenAI, Anthropic, and Google, to market their new software suites as teams of digital colleagues capable of independent action. Jensen Huang, CEO of Nvidia, has previously described this future as workplaces populated by “digital humans.” The shift is already happening in some organisations, with nearly a third of the 1,261 managers in Wiles’s study reporting that their companies frame AI agents as employees. Twenty-three per cent of those managers even list these tools on official org charts.
The cost of naming
While agentic AI has made measurable progress at handling complex loops, treating these programs as coworkers creates dangerous illusions. The research suggests that when humans view an AI as an employee, they feel less personally responsible for its output. Consequently, participants in the study were 44% more likely to escalate questionable work to a manager for review instead of correcting it themselves. This behaviour defeats the purpose of using the tool to save time.
The problem extends beyond office dynamics. As AI agents enter healthcare, warfare, education, and government, there is a risk they will become convenient scapegoats for failures rooted in bad human decisions or poor oversight. The recent bombing of a girls’ school in Iran was widely blamed on Claude, yet evidence points to a cascade of human errors. Daron Acemoglu, an economist at MIT who won the Nobel Prize in 2024, argues that framing AI as a replacement for humans is a losing proposition. He believes these systems should instead be optimised to improve human capabilities, which is not currently the case.
What workers actually want
At Stanford, researchers asked 1,500 workers across 104 jobs what tasks they would find most helpful. While law clerks wanted AI to track progress on cases, the tasks deemed most suitable by experts were often rejected by the staff. Sales representatives, for instance, did not want agents verifying customer credit ratings. Calling an AI tool Alex and giving it a title is a branding exercise that does not improve the software. It makes the humans working alongside it less effective at their own jobs.
What it means
The industry is pushing for a shift in terminology that alters how people interact with technology. Calling an AI agent a coworker invites users to abdicate responsibility and lowers the standard of work. The tools are not ready for the role, and the human operators deserve a clearer understanding of what they are actually using.
