2025 年 Stack Overflow 开发者调查(第 15 年)覆盖 177 个国家、314 种技术栈:84% 的开发者在用 AI 工具创新高,但正面评价从 70% 降到 60%,呈现“用得更多、满意更低”。最大痛点不是“完全错”,而是“差不多对但不完全对”(66%);45% 认为调试 AI 生成代码更耗时。对准确性的态度分裂:46% 不信任、33% 信任,真正“高度信任”仅 3%;35% 访问 Stack Overflow 是为了解决 AI 相关问题。
AI Agent 仍未普及:52% 不用,其中 38% 明确没计划;每天用的只有 14%。但在使用者里,69% 认为提升生产力、70% 认为缩短特定任务时间——反映“效果存在但门槛更高”,核心是信任与风险:补全代码与让其自主操作文件/API 的代价不同。实践上,开发者把 AI 视为“需管理的助手”,而不是“自动驾驶”。
技术与角色趋势:JS/TS 使用率 66%,Python 同比 +7 个百分点到 58%;PostgreSQL 使用率 55.6%。模型使用上 GPT 系列用过的占 81%,Claude Sonnet 用过 42.8% 但好评率 67.5%(高于 GPT 的 61%),呈“叫好不叫座”。72% 反对 vibe coding(另有 5% 表示绝对不),约 15% 承认在做;AI 的“禁区”集中在部署/监控(76% 不打算用)、项目规划(69%)与提交/评审(61%)。调查首次单列“架构师”,6.1% 即进前四;同时 24.5% 表示工作快乐(去年 20%),32% 全远程,65% 从业超 10 年,63.6% 认为 AI 不威胁工作。
Stack Overflow’s 2025 developer survey (15th year; 177 countries; 314 stacks) shows AI is now mainstream: 84% use AI tools, yet positive sentiment fell from 70% to 60%. The biggest complaint is “almost right, but not quite” (66%), not being totally wrong. As a result, time shifts from writing to verifying: 45% say debugging AI-written code takes longer than their own.
Trust is the limiter. Accuracy is distrusted by 46% and trusted by 33%, and only 3% report high trust. AI Agents lag even more: 52% don’t use them (38% have no plan), and just 14% use them daily—though among users, 69% report productivity gains and 70% say certain tasks take less time. Developers prefer human-in-the-loop control, especially when tools can execute sequences or touch files/APIs.
Trends concentrate around pragmatic adoption. JS/TS remains #1 (66%), while Python rose 7 points to 58%; PostgreSQL leads databases at 55.6%. Model usage is uneven: 81% have used OpenAI GPT, while Claude Sonnet is used by 42.8% yet rates higher (67.5% vs GPT’s 61%). Most reject “vibe coding” (72% + 5% “absolutely not”; ~15% admit it) and draw hard boundaries around deployment/monitoring (76% won’t use AI), planning (69%), and code submission/review (61%). A new “architect” role appears at 6.1%, reinforcing that as coding gets cheaper, system design and accountability become the value center.