EN
Key Findings
- 81% of surveyed social scientists have tried AI chatbots for research, but only 20% have adopted coding agents like Claude Code into their workflow.
- Coding agent adoption is highly uneven: researchers with male names adopt at more than twice the rate of those with female names; top university researchers are 40% more likely to use them.
- Coding agent users post more working papers and grant proposals than peers, but don't yet submit more journal papers.
- 88% of researchers are optimistic about AI raising paper productivity, but 70% are more optimistic about individual productivity than about broader field impact.
How are AI coding agents changing how we study the economy and society?
Agentic coding platforms like Claude Code and Codex can take a research idea and a dataset, write and run an analysis, interpret the output, and iterate autonomously. What had been irreducibly human steps in empirical research can, for the first time, be automated.
These tools could accelerate science and make it more daring; they could also amplify disparities in research resources and exacerbate congestion in the scholarly record.
Survey results
The survey of 1,260 quantitative social scientists (fielded February-March 2026) found:
Adoption rates by discipline:
- Economics: 39%
- Political science: 25%
- Public health: 6%
- Education: 4%
By career stage:
- Doctoral students/postdocs: ~27%
- Tenured professors: ~13%
By gender:
- Typically male names: 2x the rate of typically female names
Main use cases (among AI users):
- Generating analysis code: 97% (coding agents) / 77% (other AI)
- Editing prose
- Methods advice
- Background on prior research
- Drafting prose: only ~33%
Productivity differences
Coding agent users report starting more projects and posting around 75% more working papers than non-users in the same discipline and career stage. However, they are not yet submitting more new papers to journals or resubmitting more quickly.
Caveats
The survey respondents are heavier AI users and more optimistic than non-responders. Productivity differences are descriptive, not causal. The early adopters may have been more productive to begin with.
中文
核心发现
- 调查显示，81% 的社会科学研究者曾尝试将 AI 聊天机器人用于研究，但只有 20% 将编程智能体（如 Claude Code）真正纳入工作流程。
- 编程智能体的采用率极不均衡：男性名字的研究者采用率是女性名字的两倍多；顶尖大学研究者使用编程智能体的可能性高出 40%。
- 编程智能体用户比同行发表了更多工作论文并提交了更多资助申请，但尚未提交更多期刊论文。
- 88% 的研究者对 AI 提升论文生产力持乐观态度，但 70% 对个人生产力提升的乐观程度高于对整个领域影响的乐观程度。
AI 编程智能体如何改变我们研究经济和社会的？
像 Claude Code 和 Codex 这样的智能体编程平台，可以接收研究想法和数据集，自主编写并运行分析、解读输出并进行迭代。经验研究中原本人力无法替代的步骤，第一次可以被自动化。
这些工具可能加速科学进程、使研究更加大胆；但也可能放大研究资源的不平等，加剧学术出版拥堵。
调查发现
对 1,260 名定量社会科学研究者（2026 年 2-3 月调查）发现：
按学科分的采用率：
- 经济学：39%
- 政治学：25%
- 公共卫生：6%
- 教育学：4%
按职业阶段分：
- 博士生/博士后：约 27%
- 终身教授：约 13%
按性别分：
- 男性名字的研究者采用率是女性名字的两倍
主要用例（在 AI 用户中）：
- 生成分析代码：97%（编程智能体）/ 77%（其他 AI）
- 编辑文本
- 方法咨询
- 研究背景查询
- 起草文本：仅约 33%
生产力差异
编程智能体用户报告比同领域同学术阶段的非用户多启动约 25% 的项目、多发表约 75% 的工作论文。然而，他们尚未提交更多新期刊论文或更快地重新提交。
注意事项
调查受访者比未回应者使用 AI 更频繁、也更乐观。生产力差异是描述性的而非因果性的。早期采用者可能本来就更有生产力。