AI as 'Coworkers': Why Humanizing Agents Is a Mistake

What happened?

A study led by Emma Wiles, a professor at Boston University, has revealed a concerning effect of humanizing artificial intelligence agents in the workplace. According to the research, when managers perceive an AI tool as a 'coworker' with a name, job title, and defined responsibilities—something companies like Microsoft, OpenAI, Anthropic, and Google are already implementing—their ability to detect errors decreases significantly. Specifically, study participants overlooked 18% more failures when the work was attributed to an 'AI employee' rather than a traditional chatbot.

The experiment, detailed in an article by MIT Technology Review published on June 29, 2026, simulated a work scenario where participants supervised tasks performed by an AI agent. One group was told the AI was a chatbot; the other, that it was an 'employee' named Alex with a title and defined responsibilities. The results showed a statistically significant difference in error detection rates, suggesting that the mere label of 'coworker' reduces critical scrutiny.

Why is it important?

This finding challenges the prevailing narrative in Silicon Valley, where the integration of autonomous agents as 'digital colleagues' working side by side with humans is promoted. Major tech companies have launched tools to manage teams of AI agents, presenting them as team members with their own identity. However, evidence suggests this strategy may be counterproductive: by assigning a status similar to that of a human employee, workers place excessive trust in the machine, reducing their critical oversight.

Automation bias, widely documented in human factors literature, describes people's tendency to trust automated decisions even when they are wrong. Humanizing AI could exacerbate this bias, leading to lower accountability and potential degradation of work quality. Additionally, it raises questions about legal and ethical responsibility: if an 'AI employee' makes a mistake, who bears the consequences? Companies adopting this rhetoric must consider the risks of human teams delegating without question, especially in critical tasks such as reviewing legal documents, medical diagnoses, or financial analysis.

Historically, over-reliance on automation has had serious consequences. For example, in aviation, the 2009 Air France Flight 447 accident was partly attributed to pilots trusting the autopilot too much. In the medical field, studies have shown that radiologists can miss anomalies when an AI system does not flag them. Wiles' study suggests that humanizing AI could amplify these risks in everyday work environments.

Consequences and context

The study adds to a growing body of literature on automation biases, where people tend to trust automated decisions even when they are incorrect. Humanizing AI could exacerbate this bias, leading to lower accountability and potential degradation of work quality. Furthermore, it raises questions about legal and ethical responsibility: if an 'AI employee' makes a mistake, who bears the consequences? Companies adopting this rhetoric must consider the risks of human teams delegating without question, especially in critical tasks such as reviewing legal documents, medical diagnoses, or financial analysis.

The current context is key: Microsoft has launched Copilot with customizable agents, OpenAI offers GPTs that act as specialized assistants, Anthropic promotes Claude as a 'trustworthy colleague', and Google has integrated agents into Workspace. All these companies use language that personifies AI, calling it a 'companion' or 'colleague'. However, Wiles' study indicates that this marketing strategy could have adverse effects on productivity and accuracy.

Moreover, the phenomenon is not limited to corporate environments. In education, students interacting with humanized AI tutors might blindly trust their answers, reducing critical learning. In the financial sector, a named AI advisor could lead clients to follow recommendations without verification, increasing the risk of poor investments.

What readers should know

• It's not just a name: Calling an AI 'Alex' is not harmless; it can change user behavior. The study shows that the 'employee' label reduces oversight by 18%.
• The effect is measurable: The 18% difference in error detection is statistically significant and relevant for high-precision environments. In tasks like contract review or data validation, this margin could translate into financial losses or legal risks.
• Alternatives: Keeping AI as a 'tool' or 'assistant' without personality could foster more critical use. Some companies are already opting for interfaces that emphasize the non-human nature of the system, such as clear 'AI' labels or robotic avatars.
• Design implications: Companies should rethink how they present their AI agents to avoid excessive trust. For example, they could avoid human names and use visual indicators that remind the user they are interacting with a machine.
• Future regulation: As AI integrates into teams, regulations may emerge requiring transparency about the non-human nature of these systems. The European Union has already proposed labeling requirements for chatbots, and it is likely to extend to more sophisticated agents.
• Legal responsibility: If an 'AI employee' makes a mistake, the company could be held liable. Humanization could complicate blame attribution, as human employees might argue they trusted the 'digital colleague'.

In summary, Emma Wiles' study is a wake-up call for the tech industry. Humanizing AI agents may seem like an attractive strategy to facilitate adoption, but it carries significant risks that must be managed. Companies, designers, and regulators must work together to ensure that integrating AI into the workplace does not compromise quality, safety, or accountability.