81% of Social Scientists Use AI Tools, But Only 20% Embrace Coding Agents
A groundbreaking survey of 1,260 social scientists conducted in early 2026 sheds light on the adoption of AI tools in academic research. While 81% of respondents reported using AI chatbots for tasks such as coding and editing, only 20% have integrated coding agents like Claude Code or Codex into their workflows. These findings highlight both the promise of AI in transforming research and the uneven distribution of its benefits.
AI Chatbots Are Common, Coding Agents Lag Behind
The survey, conducted in February and March 2026, revealed that AI chatbots have become a go-to tool for many researchers. However, the adoption of coding agents—advanced tools capable of autonomously generating, executing, and iterating on analysis code—remains limited. Claude Code emerged as the most popular coding agent, used by 86% of adopters, followed by Codex at 31%.
The limited adoption of coding agents is surprising given their potential to accelerate research. These tools can automate core tasks like data analysis and hypothesis testing, which have traditionally required human intervention. Yet, even among researchers already inclined to experiment with AI, only a fifth have embraced these more advanced systems.
Adoption Gaps: Gender, Status, and Career Stage
The survey revealed striking disparities in who uses coding agents. Male researchers were more than twice as likely as their female counterparts to adopt these tools. Similarly, researchers at top universities were 40% more likely to use coding agents than those at less prestigious institutions. Early-career academics, such as doctoral students and postdocs, were the most frequent adopters, likely reflecting their higher comfort with technology and greater career pressures to publish.
Field-specific adoption rates also varied significantly. Economists led the charge, with 39% using coding agents, compared to just 6% in public health and education. These gaps suggest that access, familiarity, and discipline-specific demands play critical roles in influencing adoption.
Boosting Productivity—But With Limits
Coding agent users reported higher productivity, posting more working papers and applying for more grants than their peers. On average, these researchers started 10% more empirical projects and posted 75% more working papers. However, this productivity bump didn’t extend to journal submissions, where no significant differences were observed. This could reflect the time lag between starting a project and submitting a polished manuscript, or the possibility that coding agents are more effective for early-stage tasks than for finalizing publishable work.
Optimism About AI, But Concerns Linger
The surveyed researchers were generally optimistic about AI's potential to enhance individual productivity, with 88% expecting it to help write publishable papers. However, fewer were confident about its broader impact on the field of social sciences. Concerns about AI potentially amplifying existing inequalities and contributing to an overload of low-quality research were common.
What’s Next?
This survey marks the baseline for an ongoing study that will include randomized experiments providing researchers with access to tools like Claude Code. Future findings will further explore whether coding agents can genuinely democratize research or whether they will exacerbate disparities in academia. As AI continues to reshape research practices, understanding its nuanced impacts will be critical.
For now, the uneven adoption of coding agents underscores a broader reality: while AI tools hold immense promise, their benefits are far from evenly distributed. How institutions and policymakers address these inequities will likely shape the future of AI-enabled research.