Framing AI as 'coworker' makes humans catch 18% fewer errors, study finds

A new study from Boston University’s Emma Wiles reveals a troubling psychological effect of anthropomorphizing AI agents. When managers were told the work came from an “AI employee” named Alex instead of a simple chatbot, they caught 18% fewer errors and were 44% more likely to escalate questionable outputs to a manager rather than correct them. This inversion of accountability—where humans feel less responsible for an “employee’s” work—runs counter to the time-saving promise of AI agents. Nearly a third of 1,261 surveyed managers said their companies already frame AI as employees, with 23% listing them on org charts.

This finding carries high stakes as AI agents are deployed in healthcare, warfare, and government. MIT Nobel laureate Daron Acemoglu warns that marketing agents as human replacements is a “losing proposition,” urging optimization for human capability enhancement instead. A separate Stanford study found that workers often disagree with tech experts about which tasks AI should automate—law clerks wanted help tracking case progress, while sales reps rejected AI for verifying customer credit ratings. The real danger, Wiles argues, is using “employee” branding to create convenient blame sinks when things go wrong, as seen in the Iran bombing incident blamed on Claude despite human errors. The message: treat agents as tools, not coworkers.