9  A relational approach to agentic AI in social science research: Projects, roles, and tasks

Thomas Davidson
University of North Carolina at Chapel Hill

Abstract: This chapter analyzes the roles and tasks that agentic AI performs in research projects and the implications for both research and pedagogy. I discuss how tasks can be augmented and automated using AI, and how this can transform the roles occupied by different people. I contend that a hybrid system involving human labor and AI agents will be the most effective solution for the future of social science research.

AI usage statement: I used a voice conversation with ChatGPT to take notes when I developed the idea for this chapter. In the same conversation, I asked ChatGPT some questions about related work and for advice on how I might structure the essay. This helped to refine my idea, although I decided not to follow the structure in my write-up. I drew the diagram by hand and used ChatGPT to generate TikZ code to render it from a photo. This involved a lot of back-and-forth. The first draft of the essay was written in Word, with some light editing suggestions from Grammarly. I used Claude Code to provide instructions for converting the paper to Markdown and uploading it to GitHub. I also used Claude Code to replace parenthetical references with the correct style based on my bib file and to proofread and suggest edits for the manuscript. The latest versions of each system available in March-June 2026 were used. This model of AI usage is closest to the augmented research described below.

9.1 Introduction

The integration of generative AI into social science research is rapidly increasing in both depth and breadth. In my field, sociology, many scholars already use these technologies for tasks like planning, writing, and data analysis (Alvero et al. 2026). This essay focuses specifically on the use of AI to aid research rather than its direct methodological applications (Bail 2024; Davidson 2024; Davidson and Karell 2025), although the line between these two uses is becoming increasingly blurred. Over the past couple of years, interfaces with generative AI have shifted from back-and-forth conversations with chatbots and much copy-and-pasting to extended dialogues with agents that can access and manipulate files and perform complex operations such as executing and debugging code. Critically, these agentic capabilities can enable AI to perform an increasing number of tasks in the research process, many of which were previously impossible to automate. As agentic AI has grown in both popularity and capabilities in 2025-2026, there has been an influx of interest among academics and much debate on social media and across blogs about the implications of these technological innovations (Messing and Tucker 2026).

The goal of this note is to outline a framework for thinking through how AI is used in the research process and its implications for research projects. Specifically, I propose a relational, role- and task-based approach to contextualize the use of these tools. Social science research involves projects with one or more researchers who take on different roles to produce one or more research papers. Each person performs distinct tasks as a function of their role, such as conducting a literature review, cleaning data, and writing manuscripts. Following recent research in labor economics and reports by AI companies, I consider these tasks as the basic unit of analysis for evaluating AI capabilities vis-à-vis human labor (Brynjolfsson et al. 2025; Handa et al. 2025; Patwardhan et al. 2025). When AI is used to perform tasks, it is important to distinguish between automation—where a task is completed without human intervention—and augmentation—where AI assists but requires human intervention. Generative AI models are being used in both capacities. A report by Anthropic analyzing user logs provides evidence of a mixture of automation and augmentation in everyday usage of the chatbot Claude (Handa et al. 2025). We currently exist on a “jagged technological frontier,” where some tasks can already be completely automated, while others may at best be augmented, or may only be performed by humans (Dell’Acqua et al. 2026). In what follows, I present four stylized depictions of research projects with varying levels of AI usage, and advocate for a hybrid approach that combines the strengths of humans and AI. Figure 9.1 presents examples for each model. Building from recent work emphasizing the need to understand the relationships between humans and machines (Tsvetkova et al. 2024), actors are represented as nodes, with color indicating whether they are human or AI, and edges denote relationships, distinguishing between human-human, human-AI, and AI-AI ties. In general, the advances in AI have led to a proliferation of human-AI and AI-AI ties, some of which may supplant human-human ties that have, until recently, been central to the research endeavor.

9.2 Four stylized approaches

Four network diagrams titled Roles, tasks, and agentic AI, showing team structures as nodes (blue for humans, tan for AI) connected by lines coded human–human, human–AI, or AI–AI. Traditional teams: a PI, grad coauthor, and two RAs, all human. Augmented researcher: a single PI connected to four AI tools — coding, writing, literature review, and reviewer. Automated research: a PI connected to an AI researcher, which coordinates coding, writing, literature review, and reviewer agents. Hybrid teams: the traditional human team plus AI agents attached to individual members, with both human–AI and AI–AI connections. The progression shows AI moving from absent, to tool, to delegated researcher, to embedded team member.
Figure 9.1: Research team configurations

9.2.1 Traditional projects

I begin by considering the conventional project, which describes most research in the social sciences prior to the development of generative AI. The top-left figure of Figure 9.1 shows an example in which a principal investigator (PI), perhaps an assistant professor, leads a research team, along with a graduate student co-author and two undergraduate research assistants, who collaborate on a research project. Such a team involves relationships among people occupying different roles, and each person has allocated tasks. For example, the undergraduate students could take on research assistant roles, where they conduct a literature review and code data; the graduate student might be given a more substantial role where they implement analyses and draft parts of the manuscript; and the PI supervises the analysis and writes and edits the final draft. While there are many possible project configurations, the commonality here is that research projects involve the distribution of tasks across roles. While I have delimited the boundaries to the primary team, there are typically other people occupying additional roles, like colleagues who provide feedback on a draft manuscript and peer reviewers who critique the paper, who also contribute to the finished research product.

9.2.2 Augmented research

Artificial intelligence enables a single researcher to perform many tasks that previously would have been distributed across a research team. AI chatbots can perform tasks like conducting a literature review, coding data, writing code, and drafting and editing a manuscript. This scenario, in which I have depicted a PI interacting with AI for various tasks, is one many researchers have found themselves in over the past few years. AI can also be given roles and instructed to assist with specific tasks. As Mollick (2024) describes, this might entail prompting the AI to adopt a persona (e.g., “You are an expert in survey analysis…”), but it can simply involve a series of role-specific conversations. The AI-augmented PI is still involved in each task, producing the research by engaging in many back-and-forth conversations with AI chatbots, or perhaps using agents to partially complete some tasks.

The delegation of tasks to agents has upsides for the PI. As studies on AI and productivity have shown (Noy and Zhang 2023), AI can speed up the research process. Unlike students—who get tired or are unavailable during midterms—AI agents are available around the clock. Nonetheless, such a configuration does not necessarily entail the replacement of human labor, as AI enables more ambitious research projects for social scientists who may not have the resources to assemble a team like the one described in the previous example. While there is much debate about whether researchers should do this and the extent to which AI can be relied upon to conduct these tasks with sufficient rigor, it is now possible for a single researcher to use AI to augment their capabilities and perform many tasks that would have previously required additional human labor.

9.2.3 Automated research

Much of the recent enthusiasm about agentic AI has centered on the automation of the research process. In contrast to the augmented model, where a PI incorporates inputs from multiple AI systems to produce the product, agentic AI enables a PI to delegate this relationship management to AI itself. As shown in the diagram, a PI might provide instructions to an AI agent, which then delegates tasks to specialized sub-agents. Critically, the research project no longer involves only human-AI interactions but also a series of AI-AI relations. For example, a coding agent might implement, run, and debug analysis code; a writing agent drafts the manuscript; and a reviewer agent suggests revisions.

Until recently, these kinds of workflows were the domain of science fiction, but advances in agentic AI have enabled the automation of much of the research workflow. Recent work shows how end-to-end agentic systems can produce reasonable first drafts of quantitative social science research papers from a single prompt (Engzell and Wilmers 2026). It may take a decade or more before we understand the upper bounds of what can be automated. In a theoretical scenario known as “superintelligence,” these systems could become so capable that automated research output will exceed the quality of that produced by any individual researcher, which would have profound implications for what it means to be a social scientist. As things stand, however, most tasks automated using current AI systems would likely be improved by an augmented approach where an expert steers the model and critically evaluates and refines its outputs. Certainly, the capability to manage an agentic workflow in which sets of AI agents perform research tasks has tremendous potential, but automation also has drawbacks, and I argue that a hybrid system that combines humans and agentic AI will be the most productive avenue for the future of social science research.

9.2.4 Hybrid human-AI research teams

In contrast to augmented and automated research, in which the roles previously occupied by human participants—graduate students and undergraduate research assistants—are supplanted by AI, I contend that a hybrid approach will be the most fruitful way to incorporate AI agents into research projects. In this scenario, the same core team is present as in the conventional research project, but some tasks are delegated to AI agents. This hybrid system entails the reallocation of tasks across roles, leveraging the strengths of humans and AI, while preserving and enhancing pedagogical goals. For example, the PI and graduate student can enlist an AI agent to write code for the analysis, enabling them to conduct more thorough testing and debugging than previously possible. Coding agents could empower an undergraduate research assistant to be involved in portions of the coding that would otherwise be beyond their expertise. By using an AI agent to conduct portions of the literature review, the research team might free up time for the research assistants to spend more time annotating and validating the data. The researcher leverages the strengths of agentic AI to tackle certain tasks via augmentation or automation, while leaving others to the research team. The hybrid approach has potential to change roles rather than eliminate them. For example, the PI might have more time to engage in hands-on data analysis that would otherwise be delegated to assistants and the graduate student could take on more of the decisions previously left to the PI.

Hybrid teams have the potential to improve social science research by incorporating the strengths of human and AI participants. Agentic AI is arguably better at coding, debugging, and data checking than the vast majority of social scientists, and has the bandwidth to review more literature and conduct more computational experiments than is practical for most people. At the same time, social scientists possess expert training, including deep domain expertise, understanding of their fields, and a taste that cannot be replicated in silico, and it is necessary, for pedagogical reasons, to have roles for students and others to complete some tasks themselves, even where automation is practical. In the most optimistic scenario, hybrid teams enable the PI to direct their energies where most effective and to provide opportunities for others to train, while leveraging the strengths of AI to automate some tasks. This allows high-quality output to be efficiently produced without compromising on scientific rigor.

9.3 Agentic AI and the future of social science research

These scenarios represent configurations that are all present, to varying degrees, in contemporary social science practice. If we are to embrace generative and agentic AI, it is important to clarify how it is being incorporated into the research process. The lens of roles and tasks provides such clarification, and allows us to better understand how AI is augmenting and automating aspects of research projects. There is much work to be done to understand how to integrate these tools most effectively. I conclude this essay with some brief remarks on the risks and benefits of agentic AI and the implications for social science research.

As things stand, most tasks in social science research cannot be fully automated, but many tasks can be augmented by AI. Often, this involves multiple back-and-forth steps. For example, a research assistant performs a preliminary literature review based on some recommendations from the PI; an AI agent conducts further searching using additional sources; the assistant shares a draft with the graduate student; and the student finalizes the literature review and checks it again using another AI agent; finally, the PI reviews and edits the final draft; and potentially enlists the RA, the graduate student, or an AI for further revisions. The ideal scenario, as AI becomes more capable, is to help streamline this process, enabling human team members to avoid rote tasks and to spend more time on the challenging, creative work of social science research.

Regardless of the workflow, it is still necessary for human researchers to have oversight and to critically evaluate work produced using AI. It is now well known that AI systems hallucinate and introduce inaccuracies, and while some of the more advanced models are now much less prone to these issues than a couple of years ago, the problem still persists. And if we delegate more tasks to AI, the consequences of any such issues will become more severe. We should also be concerned with how model training might result in path dependencies, in which a model ends up producing research that is too narrow and derivative (Hao et al. 2026) or, worse, perpetuating bad science by attempting to paper over problems and p-hacking results (Messing and Tucker 2026). To identify and mitigate these issues, we must increase transparency regarding how AI is used in social science research, including protocols for recording prompts and agent activity logs using version control.

A related risk concerns the undisclosed use of AI in research projects. The incorporation of AI into research puts a premium on the tasks that can only be performed well by humans, yet AI creates a risk of defection even on these tasks. This is already a problem in online surveys, as respondents use AI rather than write something authentic for open-ended responses (Zhang et al. 2025) or use bots to complete questionnaires (Westwood 2025). In the context of research projects, an undergraduate RA might ask an AI to annotate some data rather than doing it themselves, or a busy coauthor might pass off an AI-generated summary as their own writing. Such undisclosed AI use will have serious ramifications, and will likely lead to corrections and retractions. It is critical to reach consensus over the boundaries of AI usage and to develop protocols for verification.

Beyond the practical implications for research, AI may also impact humans’ ability to complete tasks. As I emphasize above, there is a tremendous pedagogical value to research projects. Undergraduate and graduate students learn valuable skills by working on projects and performing tasks. If we delegate these tasks to AI, then we may fail to adequately train the next generation of social scientists. Even established researchers may suffer from skill atrophy as they become reliant on AI to perform tasks. While the research on this topic is nascent, there is evidence that people who use LLMs to write have lower cognitive engagement with their work and struggle to recall quotes (Kosmyna et al. 2025). Such atrophy may contribute to increased automation as humans become less capable at performing tasks, making further automation harder to resist. It will be important to identify ways to protect the pedagogical goals of research, even for tasks that can be reliably automated.

There are tremendous opportunities for agentic AI to enhance social science research by automating some tasks and augmenting others. But any application of these tools involves risks, both immediate, such as hallucinations, and long-term, if dependence results in skill atrophy. Moving forward, it is imperative that we understand AI capabilities and establish whether a given system reliably performs a given task, and regularly update these assessments as the jagged frontier shifts. But beyond capabilities alone, it is equally important to consider normative questions regarding which tasks we should automate or augment and what roles should look like. We are currently far from a consensus on these matters, but I look forward to continued experimentation, discussion, and debate as we grapple with how agentic AI is transforming social science research.