Research Agenda

We study whether access to expert-style AI feedback raises the productivity of early-career researchers. In a randomized controlled trial with more than 100 early-career economists, treated participants receive access to a structured, multi-agent AI system (ResearchAI) that delivers diagnostic feedback on the high-context evaluative work that determines publication success. We test whether the productivity effect operates through a reallocation of effort toward judgment-intensive tasks, and whether gains concentrate among researchers closer to the knowledge frontier or accrue disproportionately to those still acquiring expertise---a distinction with direct implications for whether AI democratizes or amplifies existing inequality in production.

with Almudena Sevilla (LSE) & Ivan Vicente (Universitat de Valencia) · AEA RCT Registry (AEARCTR-0017749)

This program investigates three integrated research agendas using the ResearchAI Development Suite---an AI-powered multi-agent peer review system. Agenda 1 examines how AI-based peer review quality compares to human review, testing whether AI evaluation can transmit expert judgment and reduce gender-based disparities in the peer review process. Agenda 2 conducts an audit study to test whether AI review systems reproduce or mitigate the biases documented in human peer review, including implicit gender detection and debiasing experiments. Together, these agendas address fundamental questions about fairness, quality, and the future of academic knowledge production.

Draft -- Not for Circulation (May 2026)

This RCT evaluates the causal impact of a gamified, AI-powered financial education platform---Ruta Financiera: El Legado---on the financial knowledge, attitudes, behaviors, and economic outcomes of female sex workers (FSWs) across three major Colombian cities: Bogota, Medellin, and Cali. The platform delivers five sequential learning modules covering personal budgeting, credit and debt management, basic investment, entrepreneurship, and formal business registration. We implement a two-arm individually randomized design targeting N=900 participants, measuring financial literacy scores and formal savings rates at six and twelve months post-randomization. This study provides the first experimental estimates of a digital financial education intervention targeting FSWs in a middle-income country.

Research Proposal (June 2026)

This book project provides a comprehensive technical and theoretical guide to the lifecycle of economic Randomized Control Trials applied to Artificial Intelligence in education. Synthesizing Gary Becker's human capital theory, statistical power analysis, and structural econometrics, the framework addresses the "credibility revolution" in empirical economics. Drawing on the Potential Outcomes Framework and real-world case studies (including the Polanco-Jimenez & De Witte AI RCT), the document details protocols for ensuring robustness, and expands the micro-foundations of RCTs to account for historical persistence, institutional political economy, and macroeconomic general equilibrium.

with Kristof De Witte (KU Leuven) · Book Manuscript (June 2026)

This project details the design of a cluster-randomized controlled trial to evaluate the causal impact of Mia's Macondo Adventure---a gamified, AI-powered mathematics platform---on standardized math scores among final-year elementary students. The intervention features a unique multi-user architecture: interactive AI-driven learning modules for students, progress-monitoring dashboards for teachers, aggregation tools for school directors, and macro-level performance heatmaps for local government officials. We propose a school-level cluster-randomized design across 80 schools (40 treatment, 40 control), stratified by municipality, school size, and historical academic performance.

Research Proposal (June 2026)