Dan Luu 写系统问题时，习惯先测量、再对照、最后才下结论。把「How out of date are Android devices?」放到智能体、评测和线上系统里，真正要问的是：默认做法会不会系统性失败。下面用中文整理成可执行的工程笔记，去掉原站导航和无关链接。
Conclusion
It's common knowledge that Android device tend to be more out of date than iOS devices, but what does this actually mean? Let’s look at android marketshare data to see how old devices in the wild are. The x axis of the plot below is date, and the y axis is Android marketshare. The share of all devices sums to 100% (with some artifacts because the public data Google provides is low precision).
If we look at the graph, we see a number of reverse-S shaped contours; between each pair of contours, devices get older as we go from left to right. Each contour corresponds to the release of a new android version and the associated devices running that android version. As time passes, devices on that version get older. When a device is upgraded, they’re effectively removed from one contour into a new contour and the color changes to a less outdated color.
值得单独记下的观察
- blue: current (API major version)
- bright red/white: 3 years
- Android growth is slowing down
- Android device turnover (device upgrade rate) is slowing down
- Fewer devices are receiving updates
落地时建议先做的 5 件事
- 用自己的真实负载测，而不是只用公开榜或厂商数字。
- 把评测设计成能抓到失败模式：平均分好看但尾部崩溃，仍然算失败。
- 智能体默认不会好好用测试；要写进流程，而不是写在口头规范里。
- 性能和正确性都要有基线，改模型或改语言前后必须能对比。
- 结论写成可回滚的决策：哪一版配置、哪一版评测集、谁签字。
和智能体产品怎么接
龙虾PRO做 OpenClaw 落地时，同样吃「先测量再扩面」这条纪律：技能、数字员工和网关都要有可复现评测，而不是只看一次演示通过。
本文侧重全链路风控方法论。落地时请用自身业务单据做回放验证，不要把示例阈值直接当生产策略。 相关：风控体检 · 方案资源
常见问题 FAQ
什么是AI智能系统？
「AI智能系统」可概括为：It's common knowledge that Android device tend to be more out of date than iOS devices, but what does this actually mean? Let’s look at android marketshare data to see how old devi 本文从定义、方法与实践要点展开说明。
为什么要关注AI智能系统？
关注AI智能系统，是因为它直接影响效率、风险与可复制性。文中指出：It's common knowledge that Android device tend to be more out of date than iOS devices, but what does this actually mean? Let’s look at android marketshare data to see how old devices in the wild are. The x axis of the plot below is date, an…
如何落地AI智能系统？有哪些关键步骤？
建议按以下路径推进AI智能系统：1) blue: current (API major version)；2) bright red/white: 3 years；3) Android growth is slowing down；4) Android device turnover (device upgrade rate) is slowing down；5) Fewer devices are receiving updates。细节见正文对应章节。
AI智能系统适合哪些人或团队？
AI智能系统更适合：产品/技术负责人、运营与增长团队、需要落地智能体或自动化的中小团队、关注「AI智能系统」方向的读者。若你只需要单次聊天式问答，可先读概念；若要上生产，请重点看步骤、权限与风控相关段落。
关于「Conclusion」，本文给出了什么结论？
在「Conclusion」部分，要点是：es in the wild are. The x axis of the plot below is date, and the y axis is Android marketshare. The share of all devices sums to 100% (with some artifacts because the public data Google provides is low precision). If we
关于「值得单独记下的观察」，本文给出了什么结论？
在「值得单独记下的观察」部分，要点是：tes 落地时建议先做的 5 件事 用自己的真实负载测，而不是只用公开榜或厂商数字。把评测设计成能抓到失败模式：平均分好看但尾部崩溃，仍然算失败。智能体默认不会好好用测试；要写进流程，而不是写在口头规范里。性能和正确性都要有基线，改模型或改语言前后必须能对比。结论写成可回滚的决策：哪一版配置、哪一版评测集、谁签字。 和智能体产品怎么接 龙虾PRO做 OpenClaw 落地时，同样吃「先测量再扩面」这条纪律：技能、数字员工和网关都要有可复