To understand how we can design LLMs for the future of work in a human-centered fashion, we must start by understanding the who. Who is being impacted by HCLLMs in the workforce? Is this the same group of people that these models are being designed for? How might different groups of stakeholders—both direct and indirect—be impacted differently? There are many potential stakeholders, including workers who directly interact with LLMs; employers who may be in charge of procuring the technology for their organization; shareholders who are interested in productivity or financial gains; and customers who may see the final artifact or output created from workers using LLMs.
A natural group to start with are the workers who interface and utilize these technologies as part of their work. Here, a recurring theme is a fundamental mismatch between what people actually want from LLMs and how those technologies are currently being designed. In part, this misalignment can be attributed to organizational constraints, which are more present in the work setting compared to personal use. For instance, Challapally et al. (2025)ReferenceChallapally, A., Pease, C., Raskar, R., & Chari, P. (2025). The GenAI Divide: State of AI in Business 2025. https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf found that employees frequently resort to using personal accounts to access LLMs (e.g., ChatGPT, Claude) as they found enterprise deployments fail to meet their needs or feel too restrictive. Yet, this workaround behavior is symptomatic of a deeper issue. The way in which LLMs are currently being used in the workplace is misaligned with worker priorities. For example, Shao et al. (2025)ReferenceShao, Y., Zope, H., Jiang, Y., Pei, J., Nguyen, D., Brynjolfsson, E., & Yang, D. (2025). Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the US Workforce. arXiv Preprint arXiv:2506.06576. conducted a survey with 1,500 U.S. workers to understand what tasks they want AI agents to be used for. Critically, they found that the tasks workers want to use agents for differ substantially from how tools are currently deployed and where industry funding is going. In tandem with the human-centered reasons for centering workers’ perspectives, this finding suggests that current AI development trajectories risk prioritizing displacement over augmentation. Doing so risks repeating dangerous historical patterns in which automation technologies lead to displacement without commensurate productivity gains or wage increases (Acemoglu & Restrepo, 2021ReferenceAcemoglu, D., & Restrepo, P. (2021). Tasks, Automation, and the Rise in US Wage Inequality. Econometrica, 89(5), 1973–2019.). To close this gap, we must ensure that the needs and desires of workers are incorporated into the design of these systems from the outset.
Taking into account the perspectives of employers or shareholders, the question becomes whether introducing LLMs improves the productivity and quality of work produced. What complicates this question is the “jagged” nature by which LLMs are useful . For some tasks, LLMs are particularly performant and can automate existing process; some roles will see more of a synmbiotic relationship where there is augmentation rather than replacement; and in others, LLMs are in fact not capable at performing requisite tasks at all (Mazeika et al., 2025ReferenceMazeika, M., Gatti, A., Menghini, C., Sehwag, U. M., Singhal, S., Orlovskiy, Y., Basart, S., Sharma, M., Peskoff, D., Lau, E., & others. (2025). Remote labor index: measuring AI automation of remote work. arXiv Preprint arXiv:2510.26787.). There is a parallel consideration around literacy—whether workers are well-equipped to use the technology. Harkening back to the “gulf of envisioning” discussed in HCI for HCLLMs, it is possible that LLMs may actually be useful for workers, but workers may not know how to best specify their intent or be unaware that such capabilities exist. This places a responsibility on employers, organizations, and policymakers to invest in AI literacy programs that equip workers with the conceptual and practical knowledge needed to participate in an AI-augmented workplace, rather than simply assuming adoption will follow deployment (Challapally et al., 2025ReferenceChallapally, A., Pease, C., Raskar, R., & Chari, P. (2025). The GenAI Divide: State of AI in Business 2025. https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf; Ma et al., 2025ReferenceMa, Q., Koedinger, K., & Wu, T. (2025). Not Everyone Wins with LLMs: Behavioral Patterns and Pedagogical Implications in AI-assisted Data Analysis. arXiv Preprint arXiv:2509.21890.).
Finally, another stakeholder we will highlight are customers that serve as another end-user in this setting. For instance, as LLM-based systems increasingly replace human touchpoints in domains such as customer service or healthcare, the impact on customers warrants equal consideration. The evidence here is mixed: while interacting with LLM systems offers speed and accuracy compared to human support, these interactions can also raise frustrations around the lack of empathy or potential for miscommunication (Huang et al., 2024ReferenceHuang, D., Markovitch, D. G., & Stough, R. A. (2024). Can chatbot customer service match human service agents on customer satisfaction? An investigation in the role of trust. Journal of Retailing and Consumer Services, 76, 103600.; Li et al., 2025ReferenceLi, Y., Gan, Z., & Zheng, B. (2025). How do artificial intelligence chatbots affect customer purchase? Uncovering the dual pathways of anthropomorphism on service evaluation. Information Systems Frontiers, 27(1), 283–300.). Furthermore, recent survey data shows that customers in the U.S. still prefer talking with a human representative over an LLM system (Gutierrez, 2026ReferenceGutierrez, S. (2026). Customer service trends & statistics for 2026: Why consumers still trust humans over AI. Survey Monkey. https://www.surveymonkey.com/curiosity/customer-service-statistics/). This concern is heightened in contexts that may be more high-stakes or emotionally-laden. Nonetheless, we also want to highlight that treating this question of AI usages as a binary choice between human-only and AI-only presents an inchoate view of the issue as there is meaningful middle ground that leverages the strengths of both. For instance, existing work has explored how we can use LLMs to upskill professionals to offer better human support, preserving this human “touch” in interaction while enhancing the quality of accessibility.