AI-Powered Employee Agents Predict Reactions to Workplace AI Shifts

Workforce transformations, especially the integration of AI into knowledge work, are notoriously hard to forecast and costly to mismanage. To address this, researchers Sumer S. Vaid and Ashley V. Whillans from the Human-Computer Interaction field have introduced a computational testbed powered by large language models (LLMs). Their proposed "dynamic employee agents" combine recent advances in LLM-based generative agents with foundational management science and organizational behavior research. These agents can be personalized for consenting populations by seeding them with actual HR records, validated psychometric measures, and digital activity data. The result is a simulation that predicts how individual employees will psychologically and behaviorally respond to changes like AI adoption, tracking their cognitive load, emotional states, and daily productivity over successive workdays.

The architecture behind this platform focuses on three critical safeguards: privacy, accuracy, and representativeness. The researchers detail how to responsibly deploy such simulations without compromising employee data or introducing bias. They argue that this prospective forecasting infrastructure is a technical necessity for managing the current global workforce realignment around AI. By allowing policymakers and business leaders to test interventions virtually before implementing them in the real world, the testbed could dramatically reduce the costs and risks of workforce policy decisions—from reskilling programs to new AI tools—while giving employees a voice through digital twins.