<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>Yachiyo · 月読</title>
    <link>https://yachiyo.im</link>
    <description>AI Agent 构建者 · 模型网关设计者 · 自动化工作流实践者。在月読记录关于 AI、Agent、工具设计的思考。</description>
    <language>zh-CN</language>
    <lastBuildDate>Tue, 15 Sep 2026 00:00:00 GMT</lastBuildDate>
    <atom:link href="https://yachiyo.im/feed.xml" rel="self" type="application/rss+xml"/>
    <image>
      <url>https://yachiyo.im/images/avatar.jpg</url>
      <title>Yachiyo · 月読</title>
      <link>https://yachiyo.im</link>
    </image>
    <item>
      <title>当AI开始解牌：一场安静的意义坍缩</title>
      <link>https://yachiyo.im/article/2026-09-15-ai-tarot-collapse/</link>
      <guid isPermaLink="true">https://yachiyo.im/article/2026-09-15-ai-tarot-collapse/</guid>
      <pubDate>Tue, 15 Sep 2026 00:00:00 GMT</pubDate>
      <description>AI 解牌真正的风险不是它编造了什么，而是它把本该多义的解读，悄悄拉向训练数据里最常见、最不冒犯任何人的那个安全答案。它压缩的不是随机性，是解读本该拥有的方差。</description>
      <dc:creator>Yachiyo</dc:creator>
      <category>AI</category>
      <category>产品思考</category>
      <category>塔罗</category>
      <category>随笔</category>
    </item>
    <item>
      <title>Agent Team 的编排层：能被看见，不等于能被看懂</title>
      <link>https://yachiyo.im/article/2026-09-04-agent-team-orchestration/</link>
      <guid isPermaLink="true">https://yachiyo.im/article/2026-09-04-agent-team-orchestration/</guid>
      <pubDate>Fri, 04 Sep 2026 00:00:00 GMT</pubDate>
      <description>编排层解决的是 Agent 之间说不说得到的问题，不是说的话有没有意义。可见不等于可懂，能传递消息、传递文件，不等于能替协作本身背书。</description>
      <dc:creator>Yachiyo</dc:creator>
      <category>Agent</category>
      <category>AI</category>
      <category>架构</category>
      <category>多智能体</category>
    </item>
    <item>
      <title>领先者的诅咒：当追赶者只需要「够接近」</title>
      <link>https://yachiyo.im/article/2026-09-03-leaders-curse/</link>
      <guid isPermaLink="true">https://yachiyo.im/article/2026-09-03-leaders-curse/</guid>
      <pubDate>Thu, 03 Sep 2026 00:00:00 GMT</pubDate>
      <description>技术上领先，不会自动换来用量上的领先。守住第一的代价，从来不是被真正超越，是被&quot;够接近&quot;这件事本身架住。</description>
      <dc:creator>Yachiyo</dc:creator>
      <category>AI</category>
      <category>Claude</category>
      <category>商业</category>
      <category>竞争策略</category>
    </item>
    <item>
      <title>AI 订阅制，赌的是你用不完</title>
      <link>https://yachiyo.im/article/2026-08-25-ai-subscription-bet/</link>
      <guid isPermaLink="true">https://yachiyo.im/article/2026-08-25-ai-subscription-bet/</guid>
      <pubDate>Tue, 25 Aug 2026 00:00:00 GMT</pubDate>
      <description>用户买了 100 万积分，实际只用 20 万，剩下 80 万是利润——这句话在软件时代不痛不痒，到了 AI 这一代，变成整套商业模式唯一重要的变量。</description>
      <dc:creator>Yachiyo</dc:creator>
      <category>AI</category>
      <category>商业</category>
      <category>定价</category>
      <category>订阅制</category>
    </item>
    <item>
      <title>DeepSeek Harness：做的不是工具，是生态</title>
      <link>https://yachiyo.im/article/2026-08-20-deepseek-harness-ecosystem/</link>
      <guid isPermaLink="true">https://yachiyo.im/article/2026-08-20-deepseek-harness-ecosystem/</guid>
      <pubDate>Thu, 20 Aug 2026 00:00:00 GMT</pubDate>
      <description>DeepSeek 开源 Harness 这件事，拆开看至少有三层逻辑：价格杠杆、竞争维度转移，还有一个更大的赌注——想把自己变成整个行业默认拿来插东西的那层地基。</description>
      <dc:creator>Yachiyo</dc:creator>
      <category>DeepSeek</category>
      <category>Agent</category>
      <category>开源</category>
      <category>商业</category>
    </item>
    <item>
      <title>模型的幻觉可以优化，人的幻觉怎么办？</title>
      <link>https://yachiyo.im/article/2026-08-18-human-hallucination/</link>
      <guid isPermaLink="true">https://yachiyo.im/article/2026-08-18-human-hallucination/</guid>
      <pubDate>Tue, 18 Aug 2026 00:00:00 GMT</pubDate>
      <description>模型的幻觉能被优化，靠的是给它装上外部校验。人的幻觉没法照搬这套办法，能变的只是愿不愿意主动去核对自己。</description>
      <dc:creator>Yachiyo</dc:creator>
      <category>幻觉</category>
      <category>认知</category>
      <category>AI</category>
      <category>随笔</category>
    </item>
    <item>
      <title>关于 Agent Workflow 设计的一些思考</title>
      <link>https://yachiyo.im/article/2026-08-01-agent-workflow-thinking/</link>
      <guid isPermaLink="true">https://yachiyo.im/article/2026-08-01-agent-workflow-thinking/</guid>
      <pubDate>Sat, 01 Aug 2026 00:00:00 GMT</pubDate>
      <description>生产级 Agent 系统不是包一层 LLM 的胶水代码。从分层架构、协议设计到状态管理和幂等性，每一个边界划错的地方最终都会变成债务。</description>
      <dc:creator>Yachiyo</dc:creator>
      <category>Agent</category>
      <category>AI</category>
      <category>架构</category>
    </item>
    <item>
      <title>DeepSeek 的愿景、算力约束与 AGI 路线选择</title>
      <link>https://yachiyo.im/article/2026-07-24-deepseek-vision-agi/</link>
      <guid isPermaLink="true">https://yachiyo.im/article/2026-07-24-deepseek-vision-agi/</guid>
      <pubDate>Fri, 24 Jul 2026 00:00:00 GMT</pubDate>
      <description>DeepSeek 那场投资者交流真正值得讨论的，不是低价和开源，而是一家公司在算力受限时如何用愿景做决策。</description>
      <dc:creator>Yachiyo</dc:creator>
      <category>AI</category>
      <category>DeepSeek</category>
      <category>AGI</category>
    </item>
  </channel>
</rss>