A lot has been written lately about how AI could transform our lives and impact jobs. We are awash in daily headlines about companies laying off workers and predictions that entire professions may soon disappear because of AI. Yet we also read how some of the same companies rehire people after discovering that AI could not replace the knowledge and talent they had lost (recent examples include Ford, IBM, and others). Beyond these examples, broader evidence of AI’s uneven effects on the labor market is beginning to emerge. A Harvard Business School team analyzing nearly all US job postings from 2019 through March 2025 found that after ChatGPT launched, postings for roles more susceptible to replacement through AI automation fell 13 percent, while postings for roles where AI could augment human capabilities grew 20 percent. What, then, will AI—and generative AI (GenAI) in particular—mean for state DOT workforces? 

The answer depends on a multitude of factors, including how we define AI and its capabilities. In the context of this article, AI capability refers to an AI system’s ability to successfully complete a task with little or no human intervention while meeting the required standards for accuracy, reliability, and safety. AI capabilities have advanced rapidly, and with the emergence of agentic AI, models can process complex multistep tasks, write and execute computer code, and, if enabled, take actions on their own to achieve a set goal. These capabilities increasingly extend into work traditionally performed by highly skilled, well-paid knowledge workers, including accountants, lawyers, engineers, architects, and planners. If AI is combined with robotics, its reach will also extend to tasks requiring field-based physical action, e.g., autonomous equipment for mowing, pavement marking, and other maintenance activities. Today, however, the emphasis is  primarily on knowledge work because GenAI can be deployed quickly through existing computers and software. It is increasingly becoming clear that AI systems excel in solving well-structured problems, such as extracting information from standard documents (e.g., the MUTCD, or AASHTO’s Green Book) and checking designs or contracts against DOT requirements. They are less reliable when work demands human judgment, negotiation, or the resolution of unusual and ambiguous situations. In essence, AI is increasingly capable of performing codified, rule-based work, but organizations still rely on human experience, ethical judgment, and flexibility. How, then, should we draw the line between human work and AI automation?  

A sensible approach would be to delineate those tasks for which AI is a good fit (i.e., tasks for which there are clear required standards that AI is able to consistently meet), and to design the workflows accordingly. For several reasons, however, this is easier said than done, and the benefits may be less than anticipated: 

(i) Performance requirements vary depending on the nature of the task, the risks involved, and the consequences of potential errors, and quantifying AI’s performance is not always straightforward for complex tasks.  

(ii) Human oversight requires time and expertise and may offset some of the anticipated efficiency gains.  

(iii) Accountability cannot be delegated to AI – at least for now – so agencies must determine who is responsible for reviewing and approving AI outputs.  

Moreover, privacy, cybersecurity, procurement, return on investment, transparency, legal responsibility, labor agreements, workforce readiness, and public trust also affect whether AI should be used.  

Even if all these concerns are addressed, the agency still may choose not to invest in AI tools due to concerns about excessive reliance on automation which may gradually erode the human expertise needed to identify and correct AI failures. More broadly, technical capability does not guarantee adoption. As a recent Brookings analysis explains, the costs of customizing,  integrating, validating, and maintaining AI systems can make automation economically unattractive even when it is technically feasible. These considerations help explain why automation of tasks often proceeds much more slowly than advances in AI capabilities might suggest. 

Nevertheless, state DOTs are exploring where AI can support their core missions. AI can help employees spend less time on mundane tasks and instead focus more on higher-value work. To do so effectively, they need to identify where AI can add value without compromising safety, accountability, institutional knowledge, and public trust. Some DOT functions, roles, and tasks are more amenable to AI assistance than others. A useful starting point is to measure AI exposure at the task level. AI exposure refers to the degree to which AI’s capabilities overlap with the work required to perform a task. This overlap may create potential for automation, in which AI performs most or all of the task with little human intervention, or augmentation, in which AI assists an employee who continues to exercise judgment and remains responsible for the work. Exposure does not mean that a task will necessarily be automated or that the associated job will disappear; few jobs consist entirely of tasks that AI can perform. 

By assessing individual tasks and then aggregating the results across jobs, organizational units, and agency functions, DOTs can identify promising applications, anticipate changes in roles and skill requirements, and prioritize AI investments. Our ongoing NCHRP 23-46 project, AI Integration and Workforce Transformation for State DOTs, conducted by ODU in collaboration with Noblis and Slalom, is exploring these issues through a task-level analytical framework. As part of the study, we will soon survey state DOT leaders and employees to better understand where AI is being used, how it may affect tasks, roles and processes, what skills and training are needed, and what organizational policies can support responsible AI adoption and workforce transformation.  

The boundary between human work and AI will continue to evolve as the technology advances. DOTs should therefore treat workforce planning as an ongoing process. Achieving  positive outcomes with AI deployments requires far more than purchasing AI tools. Agency policies must be informed by a clear understanding of the strengths and challenges of AI, associated costs, potential workforce impacts, training needs, adaptation options, and the time and effort required to redesign work. Workforce training and skill development was the most frequently cited enabler of AI adoption in a 2026 AASHTO survey of state DOTs. 

Just as DOT leaders need to be aware of the challenges of AI and the impact AI may have on workforce reshaping, they must also recognize its potential to improve and expand the services their agencies provide. AI can enhance the quality and broaden the range of services agencies provide, but agency leaders must remain in the driver’s seat and decide which AI use cases advance the public mission and strengthen the workforce responsible for delivering those services.  

Mecit Cetin is Professor of Civil and Environmental Engineering at Old Dominion University; Cynthia M. Tomovic is Professor of Workforce and Organizational Development at Old Dominion University.

Disclaimer: The opinions and conclusions expressed or implied are those of the authors and are not necessarily those of the Transportation Research Board (TRB) or its sponsoring agencies. This article is not an official publication of the National Cooperative Highway Research Program (NCHRP), TRB, National Research Council, or The National Academies.