NBER working paper exploring how AI agents transform economic research methodology.
The NBER working paper AI Agents for Economic Research by Anton Korinek explores how AI agents are transforming economic research methodology by enabling autonomous systems to plan, use tools, and execute multi-step research tasks. These AI agents, built on large language models (LLMs), go beyond simple chatbots by integrating planning capabilities, memory, and access to external tools, allowing them to autonomously conduct literature reviews, write and debug econometric code, fetch and analyze economic data, and coordinate complex research workflows.
The paper emphasizes that economists without programming expertise can build their own AI research assistants using "vibe coding"—programming through natural language—and modern agentic frameworks like LangGraph, enabling the creation of sophisticated tools in minutes. It provides complete, working implementations and step-by-step code to demonstrate how AI agents can be employed throughout the entire research process, from initial investigation to final analysis.
Korinek argues that the evolution of AI—from basic LLMs to reasoning models and now to autonomous agents—represents a structural transformation in how economic research is conducted. His earlier work in 2023 laid the foundation by exploring generative AI applications in economics, and by 2025, the focus had shifted to fully autonomous agents capable of executing end-to-end research tasks.
Other related research highlights specialized agentic workflows where distinct agents handle ideation, literature discovery, empirical estimation, data cleaning, model design, and manuscript proofreading, with inter-agent communication structured to mirror the economic research pipeline. Additionally, AI agents are being explored for use in survey design and data collection as synthetic respondents, raising new questions about bias and methodological innovation.
The broader economic implications of AI agents extend to digital markets, where they may reduce transaction costs by acting as autonomous market participants capable of searching, negotiating, and transacting on behalf of users, though this also introduces new frictions such as congestion and price obfuscation
This NBER working paper investigates the integration of Large Language Models (LLMs) as autonomous agents to automate and augment the workflow of economic research. Moving beyond static prompt-response interactions, the authors construct a framework where AI agents utilize tools—such as code execution environments, web search, and file systems—to perform complex, multi-step research tasks. The study details how these agents can be orchestrated to simulate the entire research lifecycle, ranging from ideation and literature review to data analysis, coding, and manuscript drafting. The technical implementation relies on agentic loops where the LLM plans a strategy, executes code, observes the output, and self-corrects, effectively functioning as a tireless research assistant capable of handling iterative debugging and hypothesis testing with minimal human intervention.
A key contribution of the paper is the empirical demonstration of these agents' capabilities, including their ability to partially replicate existing economic studies and generate novel analytical insights. The authors provide a taxonomy of agent roles—such as the "Research Manager," "Coder," and "Writer"—illustrating how specialized sub-agents can collaborate to solve complex problems. The insights suggest that while current AI agents are not yet fully autonomous researchers due to limitations like hallucinations and context window constraints, they are sufficiently advanced to serve as powerful force multipliers. They can significantly lower the marginal cost of sophisticated analysis, democratizing access to high-level economic modeling and accelerating the iteration speed of empirical work.
This material matters because it signals a paradigm shift in the methodology of social sciences, challenging the traditional manual approach to data analysis and modeling. It posits a future where the economist’s role evolves from a "coder" to a "manager of AI agents," requiring new literacy in prompt engineering and system architecture. By providing a concrete roadmap for implementing these systems, the paper serves as a foundational text for understanding the future of human-AI collaboration in academic and policy-oriented economics. It raises essential questions regarding reproducibility, peer review standards, and the changing skill set required for the next generation of economic researchers.
This NBER working paper (W33145) examines the transformative potential of AI agents in economic research, focusing on how these autonomous systems can augment or replace traditional methodological approaches. The paper explores AI agents' capabilities in data collection, hypothesis generation, experimental design, and even policy simulation, leveraging large language models (LLMs), reinforcement learning, and multi-agent systems. Key insights include the ability of AI agents to handle complex, dynamic economic environments more efficiently than human researchers, while also introducing new challenges such as interpretability, bias, and the need for robust validation frameworks. The authors highlight case studies where AI agents have successfully replicated or extended economic models, suggesting that these tools could democratize access to high-quality research by reducing barriers to entry for non-specialists.
The paper’s contributions are particularly relevant to economists and data scientists working at the intersection of AI and social science. By formalizing the role of AI agents in economic research, the authors provide a framework for evaluating their reliability, ethical implications, and integration into existing workflows. The work underscores the importance of developing standards for AI-generated economic insights, given their potential to influence policy and market decisions. For a technically literate audience, this paper serves as both a primer on the current state of AI in economics and a call to action for further research into the methodological and ethical considerations of AI-driven economic analysis.
Why It Matters: As AI agents become more sophisticated, their role in economic research could redefine how theories are tested, policies are modeled, and data is interpreted. This paper is a timely contribution to ongoing debates about the future of economics in the age of AI, offering both optimism about increased efficiency and caution regarding unintended consequences. For researchers and practitioners, it provides a roadmap for leveraging AI tools while mitigating risks like overfitting, reproducibility issues, and algorithmic bias. The implications extend beyond academia, with potential applications in regulatory bodies, financial institutions, and policy think tanks.
Source: [NBER Working Paper No. 33145](https://www.nber.org/papers/w33145)