<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>YannTalk</title><description>AI 与工程实践</description><link>https://yanntalk.com/</link><language>zh-CN</language><item><title>用了 AI Coding，我反而不知道该怎么分任务了</title><link>https://yanntalk.com/posts/ai-task-split-logic/</link><guid isPermaLink="true">https://yanntalk.com/posts/ai-task-split-logic/</guid><description>AI Coding 把单人执行效率拉上去了，但分工方式没跟着变，提速很快到天花板。按老方法分任务，对齐成本吃掉了并行的收益。</description><pubDate>Sat, 20 Jun 2026 00:00:00 GMT</pubDate><category>AI工程</category></item><item><title>十几年 Git 历史全是 add，一个 Skill 改掉了</title><link>https://yanntalk.com/posts/git-commit-habit/</link><guid isPermaLink="true">https://yanntalk.com/posts/git-commit-habit/</guid><description>commit message 十几年全写 add，知道不对但一直放着。写了一个 Skill 让 Claude 读 diff 生成 message，几分钟搞定，顺带解决了颗粒度和 AI 代码 review 的问题。</description><pubDate>Sat, 20 Jun 2026 00:00:00 GMT</pubDate><category>AI工程</category></item><item><title>SKILL.md 怎么写：不是越写越长，是越蒸越准</title><link>https://yanntalk.com/posts/skill-distillation/</link><guid isPermaLink="true">https://yanntalk.com/posts/skill-distillation/</guid><description>写 SKILL.md 多半被当文档，越写越长越写越虚。Skill 其实是反复加规则、再反复删规则蒸出来的。加是熵增 AI 能帮，删是熵减 AI 替不了。</description><pubDate>Sat, 20 Jun 2026 00:00:00 GMT</pubDate><category>AI工程</category></item><item><title>业务全 Skill 化，不叫 AI 落地</title><link>https://yanntalk.com/posts/skill-misconception/</link><guid isPermaLink="true">https://yanntalk.com/posts/skill-misconception/</guid><description>身边不少人把传统业务流程包成 Skill 交给 Agent 跑，用 Skill 数量衡量 AI 落地。但按旧流程步骤切 Skill，堆得越多，Agent 能自主判断的地方越少。</description><pubDate>Sat, 20 Jun 2026 00:00:00 GMT</pubDate><category>AI工程</category></item><item><title>SKILL.md 不是 SOP，是你的判断小抄</title><link>https://yanntalk.com/posts/skill-not-sop/</link><guid isPermaLink="true">https://yanntalk.com/posts/skill-not-sop/</guid><description>翻了不少别人写的 SKILL.md，几乎都长成 SOP。SOP 是写给会照做的人看的，Skill 是写给会推理的 AI 看的。把 SKILL.md 写成 SOP，AI 就只是个更慢的查询系统。</description><pubDate>Sat, 20 Jun 2026 00:00:00 GMT</pubDate><category>AI工程</category></item><item><title>Skill 的原理：不是工具，是给 AI 的 4 张便签</title><link>https://yanntalk.com/posts/skill-principles/</link><guid isPermaLink="true">https://yanntalk.com/posts/skill-principles/</guid><description>很多人把 Skill 当 API，看见 scripts/ 就以为是工具。其实 Skill 是给 AI 的 4 张便签：description 永远在，SKILL.md 触发才在，references 找才在，scripts 是 AI 调的。</description><pubDate>Sat, 20 Jun 2026 00:00:00 GMT</pubDate><category>AI工程</category></item><item><title>AI 做 Code Review 总说没问题，因为它在偷懒</title><link>https://yanntalk.com/posts/ai-code-review/</link><guid isPermaLink="true">https://yanntalk.com/posts/ai-code-review/</guid><description>写了个 Skill 让 Codex 帮我做 code review，审 Claude 的代码。以为两个模型分工互不偏袒就稳了。做了几个月才发现：换谁审都一样，问题在评判机制本身。</description><pubDate>Thu, 18 Jun 2026 00:00:00 GMT</pubDate><category>AI工程</category></item><item><title>让 AI 写方案，越用 Skill 反而越差</title><link>https://yanntalk.com/posts/ai-plan-skill/</link><guid isPermaLink="true">https://yanntalk.com/posts/ai-plan-skill/</guid><description>同事不用 Skill 直接让 AI 生成的 HTML 清晰得很。我自己加 SKILL.md 改三轮反而越改越差。AI 写方案这件事，第一稿别给 Skill。</description><pubDate>Mon, 15 Jun 2026 00:00:00 GMT</pubDate><category>AI工程</category></item><item><title>AI 卡住了，我问了它一句：需要我帮忙吗？</title><link>https://yanntalk.com/posts/ai-stuck/</link><guid isPermaLink="true">https://yanntalk.com/posts/ai-stuck/</guid><description>Claude Code 写 Skill 时卡在飞书懒加载的死循环里。我问了一句&quot;需要我帮忙吗？&quot;，配合它刷新浏览器，问题就解了。</description><pubDate>Wed, 10 Jun 2026 00:00:00 GMT</pubDate><category>AI协作</category></item><item><title>[贴图] 用不好 Agent，不是模型的问题</title><link>https://yanntalk.com/gallery/feynman-agent-limit/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-agent-limit/</guid><description>用费曼方法解释Agent的边界</description><pubDate>Sun, 31 May 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>AI Coding 之后，任务能不能拆，上下文说了算</title><link>https://yanntalk.com/posts/ai-task-split-method/</link><guid isPermaLink="true">https://yanntalk.com/posts/ai-task-split-method/</guid><description>上一篇说了分工标准得变，这篇讲具体怎么判断。核心就两个问题：上下文是否完整，决策权落不落在这个域内。</description><pubDate>Fri, 29 May 2026 00:00:00 GMT</pubDate><category>AI工程</category></item><item><title>[贴图] 苹果是水果还是公司？AI 靠这招瞬间判断</title><link>https://yanntalk.com/gallery/feynman-attention/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-attention/</guid><description>用费曼方法解释注意力机制</description><pubDate>Wed, 20 May 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>[贴图] 以前 AI 给步骤，现在它直接帮你做完</title><link>https://yanntalk.com/gallery/feynman-agent/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-agent/</guid><description>用费曼方法解释Agent</description><pubDate>Mon, 18 May 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>[贴图] 你问 AI 最新的事，它答的全是一年前的</title><link>https://yanntalk.com/gallery/feynman-rag/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-rag/</guid><description>用费曼方法解释检索增强生成</description><pubDate>Fri, 15 May 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>[贴图] 用AI出的东西&apos;看着还行&apos;，你会停下来再想想吗</title><link>https://yanntalk.com/gallery/taste-scarcity/</link><guid isPermaLink="true">https://yanntalk.com/gallery/taste-scarcity/</guid><description>品味正在成为最稀缺品</description><pubDate>Mon, 11 May 2026 00:00:00 GMT</pubDate><category>品味</category></item><item><title>[贴图] 你以为 AI 更聪明了，其实它只是多了一张草稿纸</title><link>https://yanntalk.com/gallery/feynman-reasoning/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-reasoning/</guid><description>用费曼方法解释Reasoning</description><pubDate>Sun, 10 May 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>[贴图] AI 每次回答不一样？因为它在抽签</title><link>https://yanntalk.com/gallery/feynman-temperature/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-temperature/</guid><description>用费曼方法解释温度参数</description><pubDate>Sun, 10 May 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>[贴图] 9.9 和 9.11，哪个大？AI 答错了——它没有真假概念</title><link>https://yanntalk.com/gallery/feynman-hallucination/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-hallucination/</guid><description>用费曼方法解释Hallucination</description><pubDate>Sun, 03 May 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>AI 放大不了你没有的</title><link>https://yanntalk.com/posts/ai-cant-amplify/</link><guid isPermaLink="true">https://yanntalk.com/posts/ai-cant-amplify/</guid><description>用 AI 折腾了半年，批量生文、视频号系列、面试辅导，全试了一遍。最后发现 AI 放大的是你已经有的东西，不是你还没有的。</description><pubDate>Fri, 01 May 2026 00:00:00 GMT</pubDate><category>AI洞察</category></item><item><title>[贴图] 加了三个字，AI 突然变成了鲁迅</title><link>https://yanntalk.com/gallery/feynman-prompt-engineering/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-prompt-engineering/</guid><description>用费曼方法解释提示词工程</description><pubDate>Mon, 27 Apr 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>[贴图] 豆包和元宝，为什么聊天风格完全不同？</title><link>https://yanntalk.com/gallery/feynman-system-prompt/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-system-prompt/</guid><description>用费曼方法解释System Prompt</description><pubDate>Thu, 23 Apr 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>LLM 是一个无状态纯函数：Harness 工程的第一条公理</title><link>https://yanntalk.com/posts/harness-01-prediction/</link><guid isPermaLink="true">https://yanntalk.com/posts/harness-01-prediction/</guid><description>新窗口里它又问了同一个问题——刚才一起磨方案的那个「它」哪儿去了？LLM 是无状态纯函数，这是 Harness 工程的第一条公理。</description><pubDate>Sun, 19 Apr 2026 00:00:00 GMT</pubDate><category>AI工程</category><category>Harness工程</category></item><item><title>[贴图] 和 AI 聊久了它会忘事？因为它的&apos;桌子&apos;满了</title><link>https://yanntalk.com/gallery/feynman-context-window/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-context-window/</guid><description>用费曼方法解释上下文窗口</description><pubDate>Sun, 19 Apr 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>AI 为什么时而神准时而乱说？大模型的六个底层机制</title><link>https://yanntalk.com/posts/feynman-ai-summary/</link><guid isPermaLink="true">https://yanntalk.com/posts/feynman-ai-summary/</guid><description>费曼讲 AI 阶段一总结。从 Token 到幻觉，用日常类比拆解大模型的六个底层机制，解释 AI 那些看起来矛盾的行为。</description><pubDate>Sat, 18 Apr 2026 00:00:00 GMT</pubDate><category>AI</category><category>费曼讲AI</category></item><item><title>[贴图] AI 说它&apos;懂&apos;你，其实只是在算距离</title><link>https://yanntalk.com/gallery/feynman-embedding/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-embedding/</guid><description>用费曼方法解释Embedding</description><pubDate>Wed, 15 Apr 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>[贴图] 百万个 Token 正在悄悄流走，就在此刻</title><link>https://yanntalk.com/gallery/feynman-token/</link><guid isPermaLink="true">https://yanntalk.com/gallery/feynman-token/</guid><description>用费曼方法解释Token</description><pubDate>Mon, 13 Apr 2026 00:00:00 GMT</pubDate><category>费曼讲AI</category></item><item><title>不知道怎么用好 AI？答案就在 AI 那里</title><link>https://yanntalk.com/posts/lixiaolai-recursive-ai/</link><guid isPermaLink="true">https://yanntalk.com/posts/lixiaolai-recursive-ai/</guid><description>李笑来分享的递归法：直接问 AI&quot;如何像专家一样使用 AI&quot;。用 AI 来解决&quot;怎么用好 AI&quot;这个问题本身——就像当年你用谷歌搜&quot;怎么用谷歌&quot;一样。</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>AI</category></item><item><title>你的 AI Coding 占比多少？面试官追问的才是重点</title><link>https://yanntalk.com/posts/ai-coding-percentage/</link><guid isPermaLink="true">https://yanntalk.com/posts/ai-coding-percentage/</guid><description>面试时问 AI Coding 占比，不是为了一个数字。数字背后的方法、踩过的坑、怎么管控质量，才是真正能看出认知深度的地方。</description><pubDate>Sun, 05 Apr 2026 00:00:00 GMT</pubDate><category>AI Coding</category><category>职业思考</category></item><item><title>Karpathy 说不需要 RAG，我试了，他是对的</title><link>https://yanntalk.com/posts/karpathy-llm-kb/</link><guid isPermaLink="true">https://yanntalk.com/posts/karpathy-llm-kb/</guid><description>个人知识库试过 Notion、Obsidian、卡片笔记法，都卡在整理。Karpathy 的 LLM 编译模式让整理这件事由 LLM 来做，实测跑了 219 篇，确实不需要 RAG。</description><pubDate>Sun, 05 Apr 2026 00:00:00 GMT</pubDate><category>AI工程</category><category>知识管理</category></item><item><title>AI 之后，工程师靠什么不被替代</title><link>https://yanntalk.com/posts/jobs-aesthetics/</link><guid isPermaLink="true">https://yanntalk.com/posts/jobs-aesthetics/</guid><description>从乔布斯斯坦福演讲里的书法课说起，聊聊 AI 时代工程师真正的壁垒不在技能，而在审美和判断力。</description><pubDate>Mon, 30 Mar 2026 00:00:00 GMT</pubDate><category>AI</category><category>职业思考</category></item><item><title>Agent 的核心代码不到 20 行，但大多数团队搭不稳</title><link>https://yanntalk.com/posts/agent-loop-harness/</link><guid isPermaLink="true">https://yanntalk.com/posts/agent-loop-harness/</guid><description>Agent Loop 的核心实现不到 20 行代码，真正的工程量在 Harness——权限控制、错误恢复、日志追踪这些外围机制。</description><pubDate>Sun, 29 Mar 2026 00:00:00 GMT</pubDate><category>AI工程</category><category>Agent</category></item><item><title>黄仁勋：AGI已实现，未来每个木匠都是程序员</title><link>https://yanntalk.com/posts/jensen-huang-interview/</link><guid isPermaLink="true">https://yanntalk.com/posts/jensen-huang-interview/</guid><description>Lex Fridman Podcast 第 494 期精读。两个半小时里，黄仁勋几乎把 AI 的现状、未来和 NVIDIA 的底牌全摊开了。最具争议的一句：&quot;我认为我们已经实现了 AGI。&quot;</description><pubDate>Sun, 29 Mar 2026 00:00:00 GMT</pubDate><category>AI</category><category>翻译解读</category></item><item><title>[贴图] 黄仁勋：AGI已实现，木匠变程序员</title><link>https://yanntalk.com/gallery/jensen-huang-interview/</link><guid isPermaLink="true">https://yanntalk.com/gallery/jensen-huang-interview/</guid><description>黄仁勋与 Lex Fridman 2.5小时访谈精华，7个核心观点</description><pubDate>Sun, 29 Mar 2026 00:00:00 GMT</pubDate><category>翻译解读</category></item><item><title>AI 越聊越蠢，不是模型不行，是你把上下文喂脏了</title><link>https://yanntalk.com/posts/context-noise/</link><guid isPermaLink="true">https://yanntalk.com/posts/context-noise/</guid><description>用 AI Coding 工具聊到一半突然犯蠢，不是上下文不够长，是上下文太脏了。工具定义、测试输出、压缩误伤，三层噪声叠加才是根因。</description><pubDate>Sun, 22 Mar 2026 00:00:00 GMT</pubDate><category>AI Coding</category><category>上下文工程</category></item><item><title>没人再聊 Prompt 工程了，但它正在变得更重要</title><link>https://yanntalk.com/posts/prompt-engineering-matters/</link><guid isPermaLink="true">https://yanntalk.com/posts/prompt-engineering-matters/</guid><description>所有人都在聊 Agent、Skill、MCP，一年前最热的 Prompt 工程突然没人提了。但它没有消失，只是换了形态，变得更重要了。</description><pubDate>Sun, 22 Mar 2026 00:00:00 GMT</pubDate><category>AI工程</category><category>Prompt</category></item><item><title>AI 能力边界：同事 3 秒破案，AI 查了半小时还猜错</title><link>https://yanntalk.com/posts/ai-intuition-gap/</link><guid isPermaLink="true">https://yanntalk.com/posts/ai-intuition-gap/</guid><description>AI 的能力边界不在任务难度，而在于那部分知识有没有被写下来。专家直觉是隐性知识，活在经历里，训练数据里根本不存在。</description><pubDate>Sun, 01 Mar 2026 00:00:00 GMT</pubDate><category>AI</category><category>AI工程</category></item><item><title>OpenClaw 创始人：编程不会消失，但会变成&quot;织毛衣&quot;</title><link>https://yanntalk.com/posts/openclaw-founder/</link><guid isPermaLink="true">https://yanntalk.com/posts/openclaw-founder/</guid><description>一个倦怠消失三年的开发者，回来后用 AI Agent 三个月造出 GitHub 24 万 stars 的开源项目。综合三场重磅访谈，提炼他关于 AI 编程、软件工程和开源未来的核心观点。</description><pubDate>Sat, 28 Feb 2026 00:00:00 GMT</pubDate><category>AI</category><category>翻译解读</category></item><item><title>[贴图] 3个月24万Stars，OpenClaw创始人的AI编程方法论</title><link>https://yanntalk.com/gallery/openclaw-founder/</link><guid isPermaLink="true">https://yanntalk.com/gallery/openclaw-founder/</guid><description>综合 Peter Steinberger 三场重磅访谈，拆解他怎么用 AI Agent 造出 OpenClaw</description><pubDate>Sat, 28 Feb 2026 00:00:00 GMT</pubDate><category>AI编程</category></item><item><title>为什么你每天教 AI，它却一个字都没记住？</title><link>https://yanntalk.com/posts/ai-remember-lessons/</link><guid isPermaLink="true">https://yanntalk.com/posts/ai-remember-lessons/</guid><description>大多数人把 AI 当一次性工具，纠正完就过了。但如果你让它把教训沉淀成规则文件，它就能越用越顺手——本质是把你的经验变成它的记忆。</description><pubDate>Mon, 23 Feb 2026 00:00:00 GMT</pubDate><category>AI工程</category><category>方法论</category></item><item><title>Agent 工程师火了，前端后端测试算法的墙正在同时倒塌</title><link>https://yanntalk.com/posts/engineer-convergence/</link><guid isPermaLink="true">https://yanntalk.com/posts/engineer-convergence/</guid><description>越来越多团队在招 Agent 工程师，前端后端测试算法的职能划分正在消失。AI Coding 工具铺开之后，按技术栈分工的逻辑不成立了。</description><pubDate>Mon, 23 Feb 2026 00:00:00 GMT</pubDate><category>AI Coding</category><category>职业思考</category></item><item><title>Obsidian 都出命令行了，背后是 AI 时代的工具逻辑巨变</title><link>https://yanntalk.com/posts/cli-comeback/</link><guid isPermaLink="true">https://yanntalk.com/posts/cli-comeback/</guid><description>越来越多的工具不是在做更漂亮的界面，而是在往回走——回到黑底白字的光标闪烁。因为&quot;用户&quot;变了，AI 不需要按钮，CLI 才是它的母语。</description><pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate><category>AI</category><category>技术洞察</category></item><item><title>用了 AI 写代码之后，我连烂代码都认不出来了</title><link>https://yanntalk.com/posts/code-atrophy/</link><guid isPermaLink="true">https://yanntalk.com/posts/code-atrophy/</guid><description>AI 写代码越来越强，但我发现自己的代码判断力在退化。不是写不出来，是连好坏都分不清了。工程师的代码衰化，比想象中来得更快。</description><pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate><category>AI工程</category><category>职业思考</category></item><item><title>别再 Vibe Coding 了，那只是 AI 编程的入门级</title><link>https://yanntalk.com/posts/direct-ai-coding/</link><guid isPermaLink="true">https://yanntalk.com/posts/direct-ai-coding/</guid><description>Vibe Coding 搞定简单场景，复杂系统需要 Agentic Engineering。而 Agentic Engineering 的核心还是软件工程：架构、监督、系统设计。</description><pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate><category>AI工程</category><category>AI Coding</category></item><item><title>刷了半年 AI 评测，不如动手做一周</title><link>https://yanntalk.com/posts/info-addiction/</link><guid isPermaLink="true">https://yanntalk.com/posts/info-addiction/</guid><description>大多数技术人的焦虑不是能力不够，而是资讯中毒——刷了一百篇 AI 文章，一行代码没写。看一百篇游泳教程，和跳进水里，是两回事。</description><pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate><category>AI</category><category>职业思考</category></item><item><title>纳瓦尔：平庸的程序员已经没有市场了</title><link>https://yanntalk.com/posts/naval-ai-motorcycle/</link><guid isPermaLink="true">https://yanntalk.com/posts/naval-ai-motorcycle/</guid><description>Naval Ravikant 播客《A Motorcycle for the Mind》精读。Steve Jobs 说电脑是&quot;思维的自行车&quot;，Naval 将其升级——AI 是头脑的摩托车，更快、更强大，但方向盘永远在人类手里。</description><pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate><category>AI</category><category>翻译解读</category></item><item><title>别再死磕 Prompt 了：三个技巧让 AI 替你写</title><link>https://yanntalk.com/posts/prompt-tips/</link><guid isPermaLink="true">https://yanntalk.com/posts/prompt-tips/</guid><description>好的 Prompt 不是你写出来的，是你和 AI 一起长出来的。给案例、让它问、让它反写，比自己绞尽脑汁高效十倍。</description><pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate><category>AI协作</category><category>方法论</category></item><item><title>硅谷教父格雷厄姆：AI 时代最值钱的不是技能，是&quot;选什么不做&quot;的判断力</title><link>https://yanntalk.com/posts/taste-scarcity/</link><guid isPermaLink="true">https://yanntalk.com/posts/taste-scarcity/</guid><description>当执行变得廉价，品味——选择做什么的判断力——成为最稀缺的能力。</description><pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate><category>AI洞察</category></item><item><title>推了多年都没人坚持的 TDD，AI 时代突然成了刚需</title><link>https://yanntalk.com/posts/tdd-comeback/</link><guid isPermaLink="true">https://yanntalk.com/posts/tdd-comeback/</guid><description>TDD 推了多年没人坚持，核心障碍是维护成本和人性。但 AI 写代码重构成常态之后，测试从可选变成了刚需——不是因为 TDD 变好了，是因为没测试的代价变大了。</description><pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate><category>AI Coding</category><category>工程实践</category></item><item><title>[贴图] 你有没有发现最近好多工具在&apos;倒退&apos;回命令行</title><link>https://yanntalk.com/gallery/cli-comeback/</link><guid isPermaLink="true">https://yanntalk.com/gallery/cli-comeback/</guid><description>CLI 回潮背后的逻辑：AI 时代的用户变了</description><pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate><category>AI工具</category></item><item><title>[贴图] 你们每天花多少时间刷AI新闻</title><link>https://yanntalk.com/gallery/info-addiction/</link><guid isPermaLink="true">https://yanntalk.com/gallery/info-addiction/</guid><description>从资讯中毒到动手实践的转变</description><pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate><category>AI焦虑</category></item><item><title>[贴图] TDD 推了十年没推动，AI 一来就成了</title><link>https://yanntalk.com/gallery/tdd-comeback/</link><guid isPermaLink="true">https://yanntalk.com/gallery/tdd-comeback/</guid><description>AI 时代测试驱动开发的回归</description><pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate><category>AI编程</category></item></channel></rss>