Calling AI Tools 'Employees' Makes Human Workers Worse at Their Jobs

The problem with AI 'coworkers'

When managers were told they'd be working with an AI "employee" named Alex rather than a chatbot, they became significantly worse at their jobs. Research by Emma Wiles, a business professor at Boston University, found that participants caught 18% fewer errors when AI output was attributed to an agentic "AI employee" compared to when the same work came from a tool explicitly labeled as software.

The study also revealed that managers felt less personally responsible for AI output when it was framed as coming from an employee. They were 44% more likely to escalate questionable work to a superior rather than trust their own corrections—defeating the entire purpose of using AI to save time.

These findings arrive as major technology companies aggressively market AI agents as digital colleagues. Since April, Microsoft, OpenAI, Anthropic, and Google have all released tools explicitly designed to manage teams of AI agents. Nvidia CEO Jensen Huang has promoted visions of workplaces filled with "digital humans." Among the 1,261 managers Wiles studied, nearly a third reported their companies already frame AI agents as employees, with 23% even placing them on organizational charts.

Why it matters

As AI agents expand into healthcare, education, government, and military applications, treating them as autonomous employees creates a dangerous accountability gap. When systems fail, the "AI employee" becomes a convenient scapegoat for what are actually human failures in decision-making, incentive structures, and oversight. This pattern already emerged when a bomb strike on a girls' school in Iran was blamed on the AI system Claude, despite evidence pointing to cascading human errors.

What workers actually want

MIT economist and Nobel laureate Daron Acemoglu argues that AI agents are being marketed as human replacements when they should instead be optimized to enhance human capabilities. Recent Stanford research supports this view: when 1,500 workers across 104 jobs were asked what AI assistance would genuinely help them, their answers often contradicted what technology experts assumed.

Law clerks wanted AI to help track progress across multiple cases. But sales representatives explicitly did not want AI agents handling tasks like verifying customer credit ratings—precisely the kind of work technologists thought most suitable for automation.

The gap between what AI companies are selling and what workers need points to a fundamental misalignment. Branding AI tools as employees may be convenient marketing, but it doesn't make the technology more capable. More importantly, it makes the humans responsible for oversight demonstrably worse at their actual jobs—the jobs that still require the agency AI is attempting to replicate.

These findings were first reported by MIT Technology Review in The Algorithm newsletter.