Cursor 团队发布《开发者习惯报告》
TL;DR · AI Summary
AI Coding 正在深刻改变软件开发形态，代码产出显著提升，AI 生成代码留存率提高，开发单位变大，Agent 工作更复杂。
Key Takeaways
- 2026 年 5 月开发者每周新增代码行数达 8.6K，比 2025 年初增长 139%。
- 1000 行以上的大 PR 占比从 8% 提升到 13.8%，AI 助力开发者处理更大任务。
- AI 生成代码留存率从 76% 提高到 81%，表明开发者更认可 AI 输出。
Outline
Jump quickly between sections.
Mindmap
See how the topics connect at a glance.
查看大纲文本（无障碍 / 无 JS 友好）
- 开发者习惯报告
Highlights
Key sentences worth saving and sharing.
代码产出速度明显提高：每位开发者每周新增代码行从 2025 年初约 3.6K，升到 2026 年 5 月的 8.6K。
1000 行以上的“大 PR”占比从约 8% 升到 13.8%。
被接受的 AI 代码在 60 分钟后仍然存在的比例，从 2026 年初约 76% 升到约 81%。
AI Coding is no longer just about "writing code faster"; it is profoundly transforming the nature of software development: developers are submitting larger changes, Agents are handling deeper tasks, more AI-generated code is being integrated into codebases and retained, and the next step will be a shift from personal assistant tools to automated development infrastructure. https://t.co/nk4ge5gStW
The most important findings from the Cursor team's report https://t.co/FIINZNXUMh" / X
Cursor has released its latest "Developer Habits Report". AI Coding is no longer just about "writing code faster"; it is significantly reshaping the landscape of software development: developers are submitting larger changes, Agents are handling more complex tasks, more AI-generated code is being integrated into codebases and retained, and the next step will be a shift from personal assistant tools to automated development infrastructure. cursor.com/insights
The 5 most important findings from the Cursor team's report:
- Code output speed has significantly increased: the average number of new lines of code added per developer per week rose from about 3.6K in early 2025 to 8.6K by May 2026; the p75 number of new lines of code in PRs also increased from about 126 to about 345.
- Development units have grown larger: the proportion of "large PRs" with over 1000 lines of code rose from about 8% to 13.8%. This indicates that AI is not just completing local code but enabling developers to handle larger tasks at once.
- Agents are performing more complex tasks: over the past two months, the average number of tool calls per Agent session increased by about 30%, showing that they are reading files, modifying files, searching code, running commands, and accessing web pages more frequently.
- The "retention" of AI-generated code is improving: the proportion of accepted AI-generated code that remains after 60 minutes increased from about 76% at the start of 2026 to about 81%. This is more meaningful than simply counting the volume of generated code, as it reflects whether the code is being accepted by developers and integrated into actual workflows.
- A small group of high-level users is gaining greater benefits: AI usage is highly concentrated, with Gini coefficients for AI-generated code lines, costs, and token usage at 0.77, 0.75, and 0.72, respectively. The P99 users generate 46 times more AI code lines and merge 15 times more PRs than median users.
Quote
Cursor
@cursor_ai
16h
Introducing the Cursor Developer Habits Report. We’re sharing some of our findings on how software development is changing. It’s based on the most comprehensive dataset on AI coding in the world, across all model families.