Abstract

Researchers conducted an experiment with 1,026 engineers in which participants evaluated a Python code snippet that was purportedly written by another engineer, either with or without AI assistance. The code itself was identical across all conditions—only the described method of creation differed. The results were striking. When reviewers believed an engineer had used AI, they rated that engineer’s competence 9% lower on average, despite reviewing identical work—and the penalty was more severe for women and older workers. This competence penalty points to a fundamental misalignment in how organizations approach AI adoption. While companies focus on access, training, and technical infrastructure, they overlook the social dynamics that determine whether employees actually use these tools.