At a glance
- What it is: The Values of Value in AI Adoption: Rethinking Efficiency in UX Designers’ Workplaces
- Who: Inha Cha, Catherine Wieczorek, and Richmond Y. Wong (Georgia Institute of Technology)
- Where: Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, April 2026, Barcelona. doi.org/10.1145/3772318.3790429
- Type: qualitative study, 5 group design workshops and 14 individual follow-up interviews.
First reading: what it does and what it finds
The question that opens the paper is not whether AI is good for design work, but how decisions to adopt it are made in practice. That distinction matters, because most earlier studies evaluate how well a tool works; this one looks at what happens in the team when the tool arrives.
To do so, the authors recruited 15 UX designers from technology companies in the United States, Canada, and South Korea. The profiles are varied: there are UX researchers at a finance firm, a senior designer in IT with 25 years of experience, a design intern, and people in consulting and health care. They organized them into five groups by sector or professional context and ran two-and-a-half-hour virtual workshops in which each participant built a hypothetical scenario of AI adoption in their own work, with exercises in Miro. Half-hour individual interviews followed. The analysis was reflexive thematic analysis: two team members worked through the material separately and then the whole team defined the themes together.
The findings are organized at three scales: individual, team, and organizational. Each reveals something different, and together they build the central argument.
At the individual level, almost all participants cited efficiency as their main reason for using AI. But on closer examination, they described what the study calls “hidden work”: time spent prompting, failed attempts, checking for hallucinations, fixing errors before having something usable. Several acknowledged that this cost is rarely measured or discussed. And behind the enthusiasm for productivity, a more persistent tension appeared: the fear of becoming dispensable. One participant described it this way: “Sometimes I wonder if I depend too much on AI… I’m automating my workflow, using AI to transcribe, pull insights from interviews, write PRDs. I feel like I’m losing my core skills.” Another, younger participant put it with a different image: you have to think about which muscles you are not using while the tool works on its own.
At the team level, the first finding is about how AI reaches the team: through word of mouth, not training. Participants learned about new tools in lunchtime conversations or from an enthusiastic colleague who became the go-to user. That informal diffusion led to uneven adoption and produced tensions: whoever adopted faster started doing tasks that others used to do (content writing, frontend implementations), and those who had those responsibilities felt displaced. The study also found a transparency problem: some team members presented AI-generated work without saying so, which eroded trust. Several participants likened the ideal solution to the logic of citing sources on Wikipedia: at some point in the process you have to declare what the person produced and what the model produced. Finally, in teams where AI already circulated freely, a pattern emerged that participants themselves called “magic eight-ball thinking”: the tendency to take the model’s output as the final answer, without subjecting it to the same debate a colleague’s proposal would receive.
At the organizational level, designers discovered that they are not the ones who decide. In most cases, formal adoption depended on managerial approval, compliance reviews, and alignment with clients. In large companies, AI existed in a “gray zone”: tools used unofficially until their value was too obvious to ignore, but approval came late, sometimes after the project. One participant described receiving compliance authorization in week 7 of an 8-week project. And while leadership measured success in terms of productivity and savings, designers described a different experience: AI did not take work off their plates; it added new responsibilities without reassigning the old ones.
The central argument is that “efficiency” is not a neutral technical descriptor. It is a contested concept. When a manager talks about efficiency, they are talking about measurable metrics; when a designer talks about efficiency, they may be talking about their ability to do meaningful work without delegating what gives them professional purpose. Those two definitions do not match, and AI adoption exposes that gap. The paper draws on the work of anthropologist David Graeber to make this precise: “value” in the economic sense (productivity, cost) and “values” in the social sense (autonomy, rigor, professional identity) are distinct registers that constantly intermingle, and that mixing is not accidental but political.
Second reading: from Latin America
The study was conducted in technology settings in the United States, Canada, and South Korea. That carries weight. UX design markets in Latin America operate with different budgets, smaller teams, and often under pressure to justify disciplines that local organizations have yet to fully value. That makes the tension between “AI makes me more efficient” and “AI makes me dispensable” potentially sharper: in small teams, automating a function can literally cost someone their job.
The informal diffusion the paper describes is probably recognizable in any design office in the region: the tool arrives through an enthusiastic colleague, not through company policy. But in organizations with more pronounced hierarchies, the room for a designer to question adoption imposed from above may be narrower than what these participants describe.
What the paper does not address, because it was not its question, is what happens when design teams in Latin America integrate AI not in the context of large technology firms but in agencies, small startup teams, or IT departments of companies that never had designers before. In those contexts, some of the organizational barriers the study describes (formal approvals, compliance reviews) simply do not exist, which can speed up adoption but also leaves no frameworks for resolving questions about who decides and who takes responsibility when something goes wrong.
What does translate directly, regardless of context, is the question the paper leaves open: before adopting AI to demonstrate modernity or cut costs, who can reject the tool without consequences? Who will absorb the hidden work when the model fails? And what happens to the people whose skills overlap with what AI starts to do? Those questions have no technical answer.
The fine print
- The study has 15 participants in technology sectors in three countries of the Global North and East Asia. Its value lies in explaining dynamics and tensions, not in measuring how frequent they are or in covering different contexts.
- The workshops used hypothetical scenarios, which makes reflection easier but can introduce distance from what happens in each team’s real day-to-day work.
- The fieldwork was carried out between May and July 2025, when language models were already mainstream but the sector was still gauging their impact on creative work. The paper centers the problem on that organizational layer (who decides, who bears the cost, who answers when something fails) rather than on the model’s capabilities; and that layer, because it involves people and institutions, changes more slowly.
Paper keywords: AI, AI Adoption, Workplace, Tech Practitioners, UX Practitioners, UX Practices
Automated reading. This text was generated by Claude, an Anthropic model, from the original source, without line-by-line human review. It may contain errors or debatable interpretations; to check any point, see the original source.