<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0"
     xmlns:atom="http://www.w3.org/2005/Atom"
     xmlns:content="http://purl.org/rss/1.0/modules/content/"
     xmlns:dc="http://purl.org/dc/elements/1.1/"
     xmlns:media="http://search.yahoo.com/mrss/">
  <channel>
    <title>2026 AI 上半场观察以及下半场预测 on 熊鑫伟（cubxxw）的简体中文博客 🇨🇳</title>
    <link>https://cubxxw.com/zh/columns/ai-2026-review-forecast/</link>
    <description>AI Builder、开源贡献者、数字游民的技术博客：Kubernetes、Go、AI 开源项目与成长思考。</description>
    <image>
      <title>熊鑫伟（cubxxw）的简体中文博客 🇨🇳</title>
      <url>https://cubxxw.com/assets/og-image.png</url>
      <link>https://cubxxw.com/</link>
    </image>
    <generator>Hugo -- gohugo.io</generator>
    <language>zh-cn</language>
    <lastBuildDate>Wed, 15 Jul 2026 09:00:00 +0800</lastBuildDate>
    <atom:link href="https://cubxxw.com/zh/columns/ai-2026-review-forecast/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>2026 下半场，AI Agent 的红海已满，蓝海在哪</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/ai-agent-red-ocean-blue-ocean-2026/</link>
      <pubDate>Wed, 15 Jul 2026 18:00:00 +0800</pubDate>
      <atom:updated>Thu, 16 Jul 2026 03:07:34 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/ai-agent-red-ocean-blue-ocean-2026/</guid>
      <description>2026 下半场，通用与水平 AI Agent（编码、客服、销售、日程）已成红海，基础模型厂商一个产品周期就能复刻。真正的蓝海在两处：垂直且承担全流程合规责任的受监管工作流，以及给 agent 造基础设施（运行时、支付、评测）。本文拆解为什么功能护城河守不住，什么才是真护城河（专有数据飞轮、领域深度、端到端责任），给出一张红蓝判据表、一份方向自查清单与一份反模式清单，并给超级个体与创业者两条不同下注，收口整个 2026 观察与预测专栏。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Solo Builder</category>
      <category domain="tag">Super Individual</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/ai-agent-red-ocean-blue-ocean-2026.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/ai-agent-red-ocean-blue-ocean-2026.jpeg" medium="image"><media:description>2026 下半场，AI Agent 的红海已满，蓝海在哪</media:description></media:content>
    </item>
    <item>
      <title>怎么才能放心把活交给一个没人盯着的 AI Agent</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/trusting-unattended-ai-agent/</link>
      <pubDate>Wed, 15 Jul 2026 16:00:00 +0800</pubDate>
      <atom:updated>Thu, 16 Jul 2026 03:07:34 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/trusting-unattended-ai-agent/</guid>
      <description>让 agent 替你干活的真正瓶颈从来不是&#34;能不能干&#34;，而是&#34;你敢不敢信它干完的&#34;。本文拆解无人值守 AI Agent 的信任三件套：执行前授权的护栏（guardrails）、可回归的评测（evals）、早晨的 HITL 复核，用复合误差这道硬数学解释为什么 95% 准确率连做 20 步只剩三成多，并把护栏按动作可逆性分成四级、给出评测集从生产事故里冷启动的具体步骤、早晨复核清单与反模式清单。信任不靠模型自觉，靠工程。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Monitoring</category>
      <category domain="tag">Security</category>
      <category domain="tag">Testing</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/trusting-unattended-ai-agent.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/trusting-unattended-ai-agent.jpeg" medium="image"><media:description>怎么才能放心把活交给一个没人盯着的 AI Agent</media:description></media:content>
    </item>
    <item>
      <title>开源模型便宜了近九成，一个人能养多大的 Agent 舰队</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/open-model-cost-collapse-agent-fleet/</link>
      <pubDate>Wed, 15 Jul 2026 14:00:00 +0800</pubDate>
      <atom:updated>Thu, 16 Jul 2026 03:07:34 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/open-model-cost-collapse-agent-fleet/</guid>
      <description>Agent 是烧 token 的工作负载，成本是它的物理定律。据测算，同等智能水平下开源大模型已便宜近九成，把一个 agent 任务从几美元打到几分钱。这一刻，个人第一次真正养得起一支 24 小时运转的 agent 舰队。本文用一笔账、一张 ASCII 图讲清 token 成本坍塌、多模型路由为何从优化变成默认架构，以及便宜背后被低估的隐藏成本。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Solo Builder</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/open-model-cost-collapse-agent-fleet.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/open-model-cost-collapse-agent-fleet.jpeg" medium="image"><media:description>开源模型便宜了近九成，一个人能养多大的 Agent 舰队</media:description></media:content>
    </item>
    <item>
      <title>AI Agent 开始&#34;反过来提示你&#34;，这一步意味着什么</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/proactive-agent-it-prompts-you/</link>
      <pubDate>Wed, 15 Jul 2026 12:00:00 +0800</pubDate>
      <atom:updated>Thu, 16 Jul 2026 03:07:34 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/proactive-agent-it-prompts-you/</guid>
      <description>这半年一个反直觉的信号是：越来越多 AI Agent 不再等你发指令，而是反过来读你的上下文、主动提示你该做什么。本文把主动式 Agent 放进&#34;命令行式→对话式→主动式&#34;的第三次交互范式迁移里，拆解它需要的持久记忆、agent 运行时与事件触发三块底座，给出一个可以直接套用的打扰预算模型与开口阈值公式，并对下半场做出预测：主动式会成为热词，但会先死一批——因为打扰成本被严重低估，信任与打扰的取舍才是真正的分水岭。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Context Engineering</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/proactive-agent-it-prompts-you.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/proactive-agent-it-prompts-you.jpeg" medium="image"><media:description>AI Agent 开始&#34;反过来提示你&#34;，这一步意味着什么</media:description></media:content>
    </item>
    <item>
      <title>让 AI 自动帮你追全网资讯，最后会卡在哪里</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/ai-auto-news-pipeline-limits/</link>
      <pubDate>Wed, 15 Jul 2026 10:00:00 +0800</pubDate>
      <atom:updated>Thu, 16 Jul 2026 03:07:34 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/ai-auto-news-pipeline-limits/</guid>
      <description>如果让一个 AI 全自动帮你追某个领域的全网资讯，它到底能走到哪一步？本文把 2026 上半场爆发的 AI 信息流水线拆开来看：订阅流、变更监控、主动检索三条技术路线各自卡在哪，三级去重能砍掉多少、砍不掉什么，以及复合误差为什么让长链条必然跑歪。access 门槛被砍到零，但产出往往停在一个更精致的收藏夹。真正的天花板是结构性的——AI 能替你知道，替不了你判断。文中给出可直接照做的搭建 checklist 与反模式清单，文末给出下半场预测。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Monitoring</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/ai-auto-news-pipeline-limits.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/ai-auto-news-pipeline-limits.jpeg" medium="image"><media:description>让 AI 自动帮你追全网资讯，最后会卡在哪里</media:description></media:content>
    </item>
  </channel>
</rss>
