Why Calling AI Agents “Coworkers” Could Backfire at Work
A growing number of companies are calling their AI tools “employees,” but new research suggests that labeling AI agents coworkers may quietly be making human workers worse at their jobs.
A study by Boston University business professor Emma Wiles, found that people caught 18% fewer errors when work was said to have come from an agentic “AI employee” rather than a chatbot, according to MIT Technology Review. The framing problem goes beyond semantics. Wiles’ research, conducted with over a thousand managers, revealed that when an AI tool was framed as an employee, participants saw themselves as less responsible for its output, and were 44% more likely to escalate its questionable work to a manager for review rather than trusting their own corrections, defeating the entire point of using an AI agent in the first place.
This is not a fringe trend either. Nearly a third of the 1,261 managers surveyed said their companies already frame AI agents as employees, with 23% even listing them on org charts. The push comes from the top. Nvidia CEO Jensen Huang has spoken about workplaces full of “digital humans,” and since April, Microsoft, OpenAI, Anthropic, and Google have all released tools for managing teams of AI agents, many explicitly marketed as digital colleagues with human like flexibility and cognitive power.
The danger of this AI agents coworkers narrative is not just about office morale. As AI tools get embedded into health care, government, and warfare, the report warns there is a growing risk of using AI as a convenient scapegoat for failures that are actually rooted in human decisions and poor oversight. The bombing of a girls’ school in Iran was popularly blamed on Claude, even though available evidence points instead to a chain of human errors.
Nobel Prize winning MIT economist Daron Acemoglu argues the marketing itself is flawed. He says AI agents are currently being marketed as replacements for humans, a losing proposition, and argues they should instead be optimized to improve human capabilities, which is not yet the case.
Interestingly, workers themselves don’t always want what tech companies assume they want. A Stanford effort presented 1,500 workers across 104 jobs with potential AI tasks and found that while law clerks wanted AI help tracking case progress, many tasks tech experts considered ideal for AI, such as verifying customer credit ratings for sales reps, were ones workers explicitly said they did not want automated.
The takeaway is simple but important for any company deploying AI agents coworkers style. Branding a tool as an employee does not make it better suited for the job, it just makes the humans working alongside it less careful and less accountable.
