SNSF Spark CRSK-1_237423

Recent advances in developing foundation models, particularly Large Language Models, are driving automation, with AI agents demonstrating capabilities that may replace human workers in certain roles, with significant economic and social consequences. New frameworks enable AI agents to collaborate autonomously in teams and hierarchies, forming AI-driven firms with AI managers and employees. Established methods allow these agents to interact directly with the economy, accelerating a shift where AI agents, powered by foundation models, increasingly shape economic and business decisions. Although still in its early stages, the evolving Agentic Economy could rapidly take shape in the coming years. Many commentators suggest that 2025 may already mark a turning point in the real-world adoption of AI agents in the economy. To ensure that research insights can meaningfully inform policy decisions regarding the development of an Agentic Economy, it is crucial to explore and establish research avenues at this early stage while its trajectory is still taking shape.

The rise of an Agentic Economy presents profound challenges for business and economics research. At its core lies the fundamental question: what happens when AI agents increasingly make economic decisions? While economics has long studied human decision-making, it is now crucial to understand how AI agents behave in economic contexts—and with what consequences. Although relevant across many sectors, we focus on AI agents’ decision-making in financial markets, where automation is advanced and autonomous agents are already in use. Moreover, financial markets offer a rich foundation for comparison, given established research on human biases, such as non-rational risk aversion shifts during downturns, herding behavior, and speculative bubbles. Studying AI agents in this context can reveal whether they mitigate, amplify, or introduce new inefficiencies and instabilities.

Our research plan builds on two key conjectures: first, that studying AI agents’ behavior and performance is inherently empirical due to the complexity of modern models; and second, that AI agents’ decision-making can be studied in laboratory settings with greater realism than human-focused experiments, which face limitations such as discrepancies between lab and field behavior and broader challenges with human-subject experiments. Building on these foundations, this project makes two key contributions. First, it develops and tests an empirical framework for systematically studying AI agents’ and AI firms’ economic decision-making, including a conceptual experimental design and the necessary technical infrastructure. Second, it applies this framework in a two-track experimental study of AI agents in financial markets: the first track examines individual AI agents making trading decisions, while the second analyzes AI agents within an AI investment firm, where decisions emerge from interactions among AI managers and AI employees. Both tracks will assess key economic behaviors, such as risk aversion shifts, herding, and speculative bubbles, under typical market conditions and during economic crises.

While the fast-evolving AI landscape presents uncertainties, the potential impact on research and policy is substantial. This project aims to equip business and economics researchers, as well as policymakers, with early insights into what could be a fundamental economic shift in the coming years.