I study how AI changes work. Most of it runs on field experiments, causal inference, and survey data. I spent eight years at the U.S. Department of Labor, mostly at the Bureau of Labor Statistics, working on the survey and estimation programs behind the national jobs and wage numbers. That is why I want to know how a measure was built before I trust what it says.

Papers and Preprints

• Scaffolding Human-AI Collaboration: A Field Experiment on Behavioral Protocols and Cognitive Reframing

  A field experiment with 388 employees at a Fortune 500 retailer, with both arms using Copilot, testing whether behavioral protocols and cognitive reframing change the quality of human-AI collaboration. Mindset training more than doubled the odds of top-quality individual output, and treatment non-compliance reached 47 percent.

  arXiv, Apr 2026 Designed and ran the experiment

• AI as Coordination-Compressing Capital

  A model of AI as capital that lowers the cost of coordination inside firms, and of how that reorganizes the boundaries and the hierarchy of the firm from the inside. The companion interactive report walks through the mechanism.

  arXiv, Feb 2026 Author Interactive report

• A Measurement-Clock Audit of O*NET Vintage in AI-Exposure Scores

  AI-exposure measures are built on O*NET occupational content, and O*NET does not update on a single clock. Task statements, survey-based ratings, and analyst-rated abilities each move on their own schedule, which biases the scores built on top of them.

  Working paper, 2026 Author Report

• Working with AI: Measuring the Applicability of Generative AI to Occupations

  A measure of how applicable generative AI is to the work activities that make up different occupations, built from real usage data rather than analyst ratings.

  arXiv, Jul 2025

• Evolving the Productivity Equation

  A macroeconomic framework for treating AI as a new factor of production. It asks what changes in standard growth accounting once digital labor enters the equation.

  arXiv, May 2025 Co-author

Microsoft Research Reports

• Microsoft Work Trend Index | Co-author
• New Future of Work Report 2025 | Co-author, Dec 2025
• New Future of Work Report 2024 | Co-author, Dec 2024
• Generative AI in Real-World Workplaces | Jul 2024
• Early LLM-based Tools for Enterprise Information Workers Likely Provide Meaningful Boosts to Productivity | Dec 2023

Open-Source Software

Code and data for the independent work are on GitHub. If you want to talk about any of this, reach me at alex@workforcefutures.net.