Demystifies autonomous LLM-based AI agents that plan, use tools, and execute multi-step tasks, with instructions for economists to build them.
AI Agents for Economic Research by Anton Korinek demystifies autonomous large language model (LLM)-based systems that can plan, use tools, and execute multi-step research tasks, providing hands-on guidance for economists to build their own agents even without programming expertise. The paper outlines how AI has evolved from simple chatbots to reasoning models and now to autonomous agents, marking a fundamental shift in how economic research is conducted. These AI agents function as semi-autonomous research assistants capable of conducting end-to-end analyses, including downloading and processing data, running econometric models, generating visualizations, and synthesizing results into coherent narratives.
A key innovation emphasized in the work is "vibe coding"—programming through natural language—which, when combined with modern agentic frameworks like LangGraph, enables economists to build sophisticated research assistants in minutes. The paper demonstrates practical applications such as automating literature reviews across hundreds of sources, writing and debugging econometric code, fetching and analyzing economic data, and coordinating complex research workflows. For instance, Deep Research agents use multi-agent architectures to process vast amounts of information in parallel, producing comprehensive reports with accurate citations within minutes—a task that traditionally took weeks.
The evolution of AI capabilities is framed in three paradigms: (1) traditional LLMs that operate like Kahneman’s "System 1" thinking—fast and intuitive but limited to single-pass responses; (2) reasoning models introduced in September 2024 that enable "System 2" deliberate, step-by-step problem solving for tasks like mathematical derivations and logical analysis; and (3) agentic systems emerging in December 2024 that combine language generation, reasoning, and autonomous action to actively investigate research questions rather than merely respond to them.
Notably, tools like Anthropic’s Claude Code, released in February 2025, exemplify the power of coding agents that generate complex software from natural language descriptions, significantly lowering barriers for economists without formal coding training. This shift is transforming the skill composition of the field, where programming proficiency is increasingly complemented by prompt engineering and conceptual modeling. Standard protocols such as MCP (Model Context Protocol) and A2A (Agent-to-Agent) are also enabling seamless collaboration across agents, signaling the emergence of a multi-agent ecosystem for economic research.
Korinek’s work builds on earlier publications, including his December 2023 Journal of Economic Literature article introducing generative AI for economists and a November 2024 update covering advances in LLM collaboration and reasoning, forming part of a semi-annual update series committed to tracking this rapidly evolving domain. By offering complete, working implementations alongside conceptual frameworks, the guide facilitates AI agent integration at every stage of the research process—from initial investigation to final analysis.
This paper provides a comprehensive technical overview of autonomous AI agents built on Large Language Model (LLM) architectures, specifically tailored for the economics community. Korinek demystifies the transition from passive chatbots to active agents capable of planning, reasoning, and executing complex, multi-step workflows. The material dissects the core components of these systems, including mechanisms for memory management, task decomposition, and tool use—such as the ability to browse the web, execute code, or query APIs. By explaining how LLMs can function as central controllers within a feedback loop, the paper illustrates how these agents can autonomously navigate unstructured problems and adapt their strategies based on intermediate results.
A significant contribution of this work is its practical, hands-on approach to agent development. Rather than limiting the discussion to theory, Korinek provides economists with actionable instructions and code frameworks for building custom agents. He outlines various architectural patterns—from simple reflex agents to more sophisticated multi-agent systems—and demonstrates how to implement them using current APIs. This guidance empowers researchers to automate labor-intensive aspects of the scientific process, such as data collection, literature synthesis, and even the drafting and replication of simulation models, effectively serving as a force multiplier for economic research.
The importance of this material lies in its potential to reshape research methodologies within the social sciences. As AI capabilities accelerate, the ability to deploy autonomous agents offers economists a powerful new lever for analyzing vast datasets and simulating complex economic environments where AI actors interact. This paper serves as a critical bridge, translating cutting-edge engineering concepts into accessible tools for economists, thereby ensuring the discipline remains at the forefront of the AI revolution. It highlights that the future of economic research will not only involve studying AI economies but actively utilizing AI agents to conduct that research.
# AI Agents for Economic Research by Anton Korinek
This paper by Anton Korinek provides a technical yet accessible introduction to autonomous AI agents—specifically those built on large language models (LLMs)—and their potential applications in economic research. The author demystifies how these agents can autonomously plan, utilize external tools, and execute multi-step tasks, bridging the gap between LLM capabilities and practical research workflows. Korinek outlines key architectural components, such as prompt engineering, tool integration, and memory management, while emphasizing the importance of alignment and safety in agent design. The paper also includes step-by-step guidance for economists to build their own AI agents, making it a valuable resource for researchers looking to leverage automation in data collection, analysis, and even hypothesis generation.
The paper’s key contributions include a framework for structuring economic research tasks as agentic workflows, a discussion of the trade-offs between different agent architectures (e.g., reactive vs. deliberative agents), and case studies demonstrating how AI agents can assist in literature reviews, dataset curation, and even policy simulation. Korinek highlights the transformative potential of these agents in accelerating research productivity while cautioning against over-reliance on unchecked autonomy. For economists and researchers in adjacent fields, this work serves as both a practical manual and a forward-looking exploration of how AI agents could reshape empirical and theoretical economics. Its relevance extends beyond academia, offering insights for policymakers and industry professionals interested in AI-driven decision support systems.