What happens when AI enters the workplace?

Generative AI can produce a dozen options in seconds. But it cannot determine which one will satisfy a client, support an organization’s goals or work within the constraints of a larger project.

Those decisions still fall to people.

A new paper co-authored by University of Michigan School of Information PhD candidates Huiran Yi and Lu Xian examines the human judgment required to make AI useful in the workplace.

The paper, “Designing with AI at Work: Designers’ Expertise and Pragmatic Decision-Making in Workplace AI Transformation,” was published in the proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency. Yi and Xian co-authored the paper with researchers Yile Zhang, Jingyan Zeng and Zifan Zhang.

Through interviews with 23 professional designers, the researchers explored how workers decide when and how to incorporate generative AI into their work. They found that although AI can generate ideas and options, designers must rely on their professional knowledge to evaluate those outputs and determine whether they are appropriate for a particular project.

“The decision is not simply about which option is the most beautiful or attractive,” Yi said. “Designers must determine which option is appropriate for the task and aligned with the client’s needs.”

The researchers describe this process as “pragmatic decision-making.” It includes interpreting client requests, understanding workplace relationships, evaluating the quality of AI-generated material and making quick decisions within the context of a larger project.

Much of that work can be difficult to see and easy to overlook.

“This sense-making is invisible because it is not written into the work task itself,” Yi said. “Designers must make decisions in the moment so that the design works for the larger project or task.”

The findings challenge the idea that introducing AI into a workplace is primarily a matter of automation or increased productivity. Instead, the researchers show that AI adoption relies on workers performing additional labor to turn the technology’s capabilities into useful, responsible and relevant work.

That labor can become particularly challenging when employers provide little guidance. Some participants encountered vague workplace policies about AI, while others worked for companies with no clear policies at all. Workers were left to interpret whether they were allowed to use AI, which tools were appropriate and how much responsibility they would bear if something went wrong.

“Adopting AI will necessarily involve trial and error, but workers bear the risk and responsibility for those experiments,” Xian said. “This transitional phase is not accounted for within current workplace structures.”

The researchers argue that companies need clearer guidance, more specialized AI training and greater recognition of the expertise workers contribute to AI-assisted projects. A generic AI course may not adequately prepare employees whose roles, responsibilities and professional standards differ significantly.

The paper also broadens conversations about AI transparency. Discussions of transparency often focus on how models are trained or how they produce results. Yi and Xian, instead, discuss how AI is embedded in production work and suggest that making visible where and how human expertise enters the production process is instrumental to AI accountability for organizations. 

“Visibility into how human expertise contributes to AI-assisted work can help companies identify which skills to hire for and where to invest in employee upskilling,” Xian said. “This can also help designers to identify which skills remain essential and which new skills to develop as AI transforms the workforce.”

Although the study focuses on designers, its implications extend to other forms of creative and knowledge work. As AI becomes embedded in more professions, understanding the future of work will require looking beyond what the technology produces to recognize the people who make its outputs valuable.

“What matters in an AI-enabled workplace is not only how productive a person is,” Yi said. “What also matters is the human expertise that cannot be replaced by AI.”

RELATED 

Check out UMSI’s PhD in Information program. 

Read more UMSI research by subscribing to our free research roundup newsletter!