When AI Becomes Your Coworker
Written by Eleanor McNamara
Edited by Gesi Huang
AI may be the most helpful coworker you’ve ever had—but left unchecked, it can quickly become the most dangerous. At its best, AI has the potential to make decisions faster and more accurately than humans can on their own. Mammography screening offers a clear example. In 2025, a study focused on mammography screening found that when AI-assisted software was implemented, there was a 29% increase in cancer detection compared to the control group, without increasing false positives. Additionally, the software resulted in a 44% reduction in the screen-reading workload for radiologists and allowed for increased detection of both small, lymph-node negative, invasive cancer, and high-grade in situ cancers (Hernstrom et al., 2025).
Yet, the same qualities that make AI so useful can also cause overreliance on it, leading people to think less critically and make high-risk mistakes. Only two years earlier, a study published in Radiology revealed how automation bias can affect radiologists’ ability to correctly read mammograms. When an AI tool suggested the wrong answer, inexperienced radiologists read only about 20% of mammograms correctly, in sharp contrast from approximately 80% when AI was right. Even very experienced radiologists, with about 10.8 years on the job, dropped from 82% to about 46% (Dratsch et al., 2023).
This issue is relevant across multiple fields. According to Gallup, “as of May 2026, 15% of U.S. employees use AI in their role, and of those users, 30% use it a few times a week or more” (Gallup, 2026). AI is becoming a coworker, and like any new hire, it is shifting the way that workplaces operate. For AI to genuinely be beneficial, it must be kept in check by workers and used as a tool, rather than becoming a replacement.
In fact, in situations where workers stay in charge, it has been shown that AI can be leveraged to improve efficiency in the workplace. In a study based on over 5,000 customer support agents, access to AI tools led to a 14% increase in productivity on average, with larger benefit for novice and low-skilled workers. (Brynjolfsson et al., 2023). Crucially, the AI only suggested how agents should respond to customers, and was “designed to augment, rather than replace, human agents” (Brunjolfsson et al., 2023). The agents had full decision-making capacity over incorporation of AI suggestions.
However, there is an important caveat. A study conducted by Boston Consulting Group (BCG) and Harvard Business School coined the term “jagged technological frontier” to describe how AI lacks the capability to cover all skills. In this study, they found that AI usage was only beneficial to productivity when used for “realistic knowledge tasks.” When workers used AI for more complex, knowledge-intensive or managerial tasks outside of this frontier, it was not reliable and even harmed human performance. AI boosted performance on tasks it was suited for while hurting performance on tasks outside its scope. Ultimately, this leaves it up to workers to identify whether AI is harming or helping the task at hand.
When people lean too heavily on AI, they risk losing the very skills they fought so hard to earn. After AI tools were introduced at four colonoscopy centers in Poland, the average rate of adenoma detection for non-AI assisted colonoscopies decreased from 28.4% before AI exposure to 22.4% after AI exposure (Budzyn et al., 2025). Because the doctors had become accustomed to AI, they had lost some of their ability to make their own independent conclusions, posing a real risk to patients. Similarly, a Microsoft and Carnegie Mellon survey of 319 workers found that those with more confidence in AI applied less critical thinking to its output, while those with more confidence in their own abilities applied more (Lee et al., 2025).
The potential of AI as a tool is enormous, but only when it is used correctly. As AI becomes a permanent part of the workplace, the workers who reap the benefits will be the ones who keep thinking critically rather than taking AI’s answers as gospel.
(Eleanor McNamara ‘28 is in the College of Industrial and Labor Relations. She can be reached ecm259@cornell.edu.)
