Abstract
Modern finance is undergoing a radical temporal shift: the transition from "institutional time" - defined by business days and settlement delays - to "machine speed," where $27 trillion in annual stablecoin volume moves with near-instant finality. This thesis explores a fundamental sociotechnical crisis: as we build AI agents to empower individual users, we are deploying them on an infrastructure that has removed the "temporal buffers" (like T+2 settlement) that historically prevented synchronized market collapses. My work addresses the core paradox of this new era: tools designed to provide micro-level clarity can inadvertently generate macro-level chaos, creating a "synchronized fragility" that operates faster than human intervention can possibly respond.
My technical project, Finance Buddy: Simulation-first AI Agents for Transparent Personal Finance Optimization, addresses the "black box" nature of current robo-advisory platforms which often erode user trust through opacity. Built as a production-ready system using Next.js, Django, and the Plaid API, Finance Buddy introduces a "simulation-first" architecture. By allowing users to preview investment outcomes and "what-if" scenarios in a risk-free sandbox before execution, the platform transforms explainability from a passive compliance requirement into an active educational tool. The system ensures that every AI-driven recommendation is traceable to verified transaction history, providing users with a rare degree of agency in an increasingly automated landscape.
However, my STS research paper, “Temporal Herding at Machine Speed: Programmable Money, Real-time Settlement, and the Emergent Risks of Agentic Personal Finance”, reveals that individual transparency is insufficient for systemic safety. Utilizing Kean Birch’s framework of "automated neoliberalism," I investigate the "visibility paradox": an AI agent can explain a trade perfectly to one user while remaining blind to the fact that it is triggering an identical, simultaneous move for millions of others. Through comparative infrastructure mapping and agent-based modeling, I demonstrate how the elimination of settlement friction redistributes power toward those who control algorithmic speed, while exposing gig workers and retail investors to "temporal herding." My research argues that for AI finance to be truly responsible, we must move beyond individual "informed consent" and design systemic "temporal circuit breakers" that preserve human-scale deliberation.
I am deeply grateful to my STS advisor, Professor Coleen Carrigan, for her guidance in articulating the ethical implications of algorithmic speed, and to Professor Rossane Vrugtman for her technical mentorship. Finally, I thank the UVA Foundry for providing the resources to prove that consumer financial AI can be powerful, explainable, and - most importantly - safe.
