The VP of engineering at a leading technology company stared at the quarterly adoption metrics with growing frustration. Twelve months after rolling out a state-of-the-art AI coding assistant—a tool that promised to boost developer productivity significantly—only 41% of engineers had even tried it. More troubling still: Female engineers were adopting at just 31%, and engineers 40 and older were adopting at 39%. This finding emerged from our research with 28,698 software engineers at the company.
The leadership team was surprised. They’d put a lot of money and thought into this. They’d invested in cutting-edge tools. They’d thoughtfully addressed access, infrastructure, and skill development. They’d assigned a dedicated team for deployment and promotion. They were ready to see productivity soar.
The company isn’t alone—across corporate America, the pattern repeats. According to a survey by Pew Research Center, two years after ChatGPT’s launch, only 16% of American workers use AI for work—despite 91% being allowed to. The conventional explanation points to skill gaps or training friction. Some employees might not recognize where AI could help them, or they lack confidence in their technical abilities. Indeed, frontline tech workers like software engineers and data scientists are more likely to use AI than workers in other industries. Yet even among these early adopters, usage remains surprisingly low and unequal. What explains this?
To understand why, we conducted a pre-registered experiment with 1,026 engineers from the same company. The design was simple: 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.
To continue reading this article, click here.
