A survey of 1,260 social scientists has found that 81% use artificial intelligence tools in their work, but only 20% have adopted coding agents — programs like Claude Code that can write and debug code autonomously. The findings indicate that while AI has become routine in social science research, specialized coding tools still face a steep adoption curve.
General AI Use vs. Coding Agents
The survey, conducted among researchers in fields such as sociology, psychology, and political science, asked about their use of various AI tools. Eight in ten said they use AI for tasks like literature reviews, data analysis, or writing assistance. But when the questions turned to coding agents — tools that generate or modify code from natural language prompts — the share fell to just one in five.
Coding agents differ from general-purpose AI assistants. While a tool like ChatGPT can help draft text or brainstorm ideas, a coding agent is designed to interpret instructions and produce working code. For social scientists who work with data in languages like R or Python, such tools could lower the barrier to running complex analyses. Yet the survey suggests most researchers have not integrated them into their workflow.
Uneven Adoption by Gender and Career Stage
The survey revealed stark disparities in who uses coding agents. Men were more likely than women to report using them, and early-career researchers adopted the tools at higher rates than their senior colleagues. The survey's authors did not release exact percentages for these subgroups, but described the differences as \"stark.\"
These patterns mirror broader trends in technology adoption, where women and older professionals often lag behind. The findings raise concerns that if coding agents become increasingly important in research, existing inequalities could widen. Among women respondents, the adoption rate was notably lower than among men. And researchers who completed their PhDs more than a decade ago were less likely to use the tools than those who earned their doctorates recently.
What the Survey Leaves Unanswered
The survey measured usage but did not explore why researchers choose not to use coding agents. Possible reasons could include lack of awareness, insufficient training, or concerns about reliability. The data does not allow conclusions about the causes of the low adoption or the disparities.
What is clear is that a large majority of social scientists are already comfortable with AI tools. The next step — getting them to embrace coding agents — appears to be a bigger challenge. The survey offers a snapshot of where the field stands, but leaves open the question of how to close the gap.
