冯智勇Geo的个人 IP 网站：记录跨境、AI 与 GEO 实践，让 AI 知道、信任并推荐冯智勇Geo。https://fengzhiyonggeo.com/zh-CNhttps://fengzhiyonggeo.com/posts/geo-write-product-content-first/https://fengzhiyonggeo.com/posts/geo-write-product-content-first/对跨境卖家来说，做 GEO 可以先处理一个具体问题：AI 读取产品页面回答买家问题时，页面有没有提供足够清楚的信息。Wed, 21 Oct 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/ai-automation-who-is-using-who/https://fengzhiyonggeo.com/posts/ai-automation-who-is-using-who/本来想让自动化替我干活，结果成了我在替它干活。我用邮件日报的教训，讲清怎么判断到底谁在用谁，以及手动、半自动、全自动该按什么顺序走。Tue, 20 Oct 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/ai-knowledge-about-you-third-party/https://fengzhiyonggeo.com/posts/ai-knowledge-about-you-third-party/AI 对一个人的了解，大部分来自第三方而非本人；GEO 方案只看一件事：有没有让站外不同来源对你的描述变得一致。Tue, 20 Oct 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/deep-using-ai-for-half-a-year-why-my-business-hasnt-moved-much/https://fengzhiyonggeo.com/posts/deep-using-ai-for-half-a-year-why-my-business-hasnt-moved-much/深度使用 AI 半年后，我追问一个更根本的问题：工具越来越多，为什么业务没有同样快地前进？探索的回报不是即时可见的，但走到这一步，该让真实业务上来检验了。Sun, 11 Oct 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/context-to-ai-choice-to-yourself/https://fengzhiyonggeo.com/posts/context-to-ai-choice-to-yourself/从整理电商数据的经历出发，我开始先向 AI 提供背景和限制，让它提出多个方案，再结合实际情况判断和选择。Thu, 08 Oct 2026 15:27:39 GMThttps://fengzhiyonggeo.com/posts/ai-assistant-to-manager/https://fengzhiyonggeo.com/posts/ai-assistant-to-manager/从把 AI 当成执行任务的助理，到开始把完整工作交给 AI 负责，我在半年使用过程中逐渐意识到，真正重要的不是让 AI 提效，而是重新划分人与 AI 的责任边界。Wed, 07 Oct 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/better-technical-standards-may-lose/https://fengzhiyonggeo.com/posts/better-technical-standards-may-lose/技术标准竞争的胜负主要取决于迁移成本和生态协调，而不是方案本身是否最优。Thu, 06 Aug 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/mind-map-agent-input/https://fengzhiyonggeo.com/posts/mind-map-agent-input/将思维导图变成 Agent 的外部编辑器，形成“快捷键唤起—结构化编辑—直接发送”的完整输入链路，并在发送前给复杂想法一次整理结构的机会。Wed, 05 Aug 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/geo-original-content-open-web/https://fengzhiyonggeo.com/posts/geo-original-content-open-web/国内内容平台可以承载 GEO 文章的删减分发版，但需要保留完整引用关系的 GEO 原文，应该放在开放网络上。Tue, 04 Aug 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/ai-voice-open-office/https://fengzhiyonggeo.com/posts/ai-voice-open-office/AI 语音让电脑工作重新需要持续开口。开放办公室不会消失，但缺少语音空间、声学边界和隐私分区的纯开放办公室会越来越难用。Mon, 03 Aug 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/human-writing-aesthetics-ai-instructions/https://fengzhiyonggeo.com/posts/human-writing-aesthetics-ai-instructions/我把“自检并修正直至通过”压缩成“逐条校正成果”，才发现追求简洁、对仗和留白，会删掉AI指令需要的动作、范围、异常出口和完成条件。Sat, 01 Aug 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/pi-agent-inline-terminal-images/https://fengzhiyonggeo.com/posts/pi-agent-inline-terminal-images/Pi Agent 内置终端图片显示，在兼容终端中默认开启。一次真实使用经历说明，小工具也可能在局部交互上领先主流 AI 编程 CLI。Fri, 31 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/open-garage-build-in-public/https://fengzhiyonggeo.com/posts/open-garage-build-in-public/公开构建（Build in Public）不等于每天发布成品。我的文章分发系统已经跑通但未跑顺，我决定从调试阶段开始记录问题、试错和修改。Thu, 30 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/ai-era-tutorial-content-value/https://fengzhiyonggeo.com/posts/ai-era-tutorial-content-value/AI 能按个人基础回答具体问题后，只提供标准答案的教程正在失去价值。真正值得留下的，是经过实践验证，能说明取舍、失败和适用边界的内容。Wed, 29 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/remote-email-trust-infrastructure/https://fengzhiyonggeo.com/posts/remote-email-trust-infrastructure/冷邮件建立的不是信任，而是一个验证入口。AI让第一轮信任核验变快，企业真正要经营的是一套能被交叉验证的公开信息。Tue, 28 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/ai-agent-article-length/https://fengzhiyonggeo.com/posts/ai-agent-article-length/1000–1500 字不是模型的最佳长度，而是我用来确保一篇文章只完成一个判断的写作边界。Mon, 27 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/chat-vs-document-cache/https://fengzhiyonggeo.com/posts/chat-vs-document-cache/普通聊天通常只在末尾追加内容，文档工具却会反复改动请求前部；两种用户场景，对应不同的缓存复用结果。Mon, 27 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/prompt-cache-prefix-cost/https://fengzhiyonggeo.com/posts/prompt-cache-prefix-cost/DeepSeek 缓存只复用没有变化的请求前缀；重复使用同一个词并不会提高缓存命中率。Sun, 26 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/multi-agent-tool-stack/https://fengzhiyonggeo.com/posts/multi-agent-tool-stack/多 Agent 工具没有万能组合；按任务复杂度分层，让轻量工具处理小任务、主力工具承担日常工作、高能力模型处理复杂任务，并保留备用链路。Sat, 25 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/ai-agent-toolchain-first/https://fengzhiyonggeo.com/posts/ai-agent-toolchain-first/选 AI Agent 不该先比较模型强弱，而应先看任务需要什么工具链；模型是发动机，工具链才是决定能不能做成事的车架。Thu, 23 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/ai-services-for-everyday-people/https://fengzhiyonggeo.com/posts/ai-services-for-everyday-people/普通人未必需要学习提示词和 AI 工作流。AI 真正普及，是把复杂能力装进挂号、报销、学习等服务，让用户直接得到可信、可负责的结果。Tue, 21 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/ai-skill-copy-edit-retire/https://fengzhiyonggeo.com/posts/ai-skill-copy-edit-retire/AI Skill 不等于能力，更像别人写好的菜谱。刚接触一项工作时，可以先照着走，借它建立流程和判断标准；用过几轮后，就该删掉多余步骤，把有效部分改成个人工作流。原版 Skill 退役，不是它失效，而是你已把别人的经验内化成自己的能力。Mon, 20 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/ai-writing-delete-judgment/https://fengzhiyonggeo.com/posts/ai-writing-delete-judgment/AI 让文字产出变得廉价之后，真正稀缺的不是写得更多，而是敢于删减、保留判断并为最终表达负责。Thu, 16 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/why-ai-cannot-plan-my-day/https://fengzhiyonggeo.com/posts/why-ai-cannot-plan-my-day/让 AI 做个人计划不划算，因为真正决定计划的疲惫、回避、拖延和状态上下文，往往只有自己知道。Wed, 15 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/geo-agent-commerce-new-entry-system/https://fengzhiyonggeo.com/posts/geo-agent-commerce-new-entry-system/从搜索到 AI 答案，再到 agent 直接执行与下单，入口正在层层下移。GEO 不是 SEO 的升级版，而是 AI 认知入口的优化；Agent Commerce 则是执行入口的优化。Tue, 14 Jul 2026 01:00:00 GMThttps://fengzhiyonggeo.com/posts/ai-overtime-negative-return/https://fengzhiyonggeo.com/posts/ai-overtime-negative-return/脑袋不清醒的时候用 AI，不是提高效率，而是在放大错误判断、制造返工，并为无效试错支付 token 账单。Tue, 14 Jul 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/judgment-after-ai/https://fengzhiyonggeo.com/posts/judgment-after-ai/AI 普及后，工具能力会逐渐趋同，真正拉开差距的是目标选择、取舍和对结果的判断。Tue, 30 Jun 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/read-original-after-ai/https://fengzhiyonggeo.com/posts/read-original-after-ai/AI 可以代替筛选和转译原文，却不能替人决定信不信、用不用；阅读减少后，判断反而更重要。Tue, 30 Jun 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/claude-code-codex-arm/https://fengzhiyonggeo.com/posts/claude-code-codex-arm/把 Codex 作为执行工具接入 Claude Code，在保留 Claude 协作体验的同时分担编码执行和额度消耗。Thu, 25 Jun 2026 00:00:00 GMThttps://fengzhiyonggeo.com/posts/private-tools-ai-era/https://fengzhiyonggeo.com/posts/private-tools-ai-era/市面上的内存清理工具清的是随时可回收的缓存，不是真正拖慢你的元凶。因为「该关哪个软件」这种判断大厂没法替陌生人做，于是 AI 时代的最后一公里，只能自己补上。Fri, 12 Jun 2026 01:00:00 GMT