<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel>
<title>MEDIA OS</title><link>https://guzhaolin.art/blog</link><description>MEDIA OS 博客，收录 Lingo 关于 AI Agent、代码评测、上下文工程、中文写作和支付产品的文章。</description><language>zh-CN</language>
<atom:link href="https://guzhaolin.art/rss.xml" rel="self" type="application/rss+xml"/>
<item><title>CLI Agent 的长任务为什么越做越不可靠</title><link>https://guzhaolin.art/blog/cli-agent-long-task-reliability</link><guid isPermaLink="true">https://guzhaolin.art/blog/cli-agent-long-task-reliability</guid><description>CLI Agent 在长任务中失控，不是模型参数逐步变差，而是当前上下文偏离了项目的可验证状态。缩短未验证跨度，比单纯扩大窗口更重要。</description><pubDate>Sun, 12 Jul 2026 16:00:00 GMT</pubDate></item>
<item><title>中文“AI 味”是怎样被训练出来的？</title><link>https://guzhaolin.art/blog/chinese-ai-prose-style</link><guid isPermaLink="true">https://guzhaolin.art/blog/chinese-ai-prose-style</guid><description>为什么多种模型会反复写出相似的中文？本文从训练数据和偏好选择追溯原因，也说明标注外包、跨模型蒸馏的公开证据到哪里为止。</description><pubDate>Tue, 07 Jul 2026 16:00:00 GMT</pubDate></item>
<item><title>Skill 到底有没有用？一次没有提升的编程基准实验</title><link>https://guzhaolin.art/blog/agent-skill-evaluation</link><guid isPermaLink="true">https://guzhaolin.art/blog/agent-skill-evaluation</guid><description>公开 Skill benchmark 的结论并不一致。本文对照 SkillsBench、SWE-Skills-Bench 和困难编程任务实测，说明没有提升时应怎样解释结果。</description><pubDate>Thu, 02 Jul 2026 16:00:00 GMT</pubDate></item>
<item><title>Coding Agent 分数上涨，测到的究竟是什么？</title><link>https://guzhaolin.art/blog/coding-agent-benchmark-boundary</link><guid isPermaLink="true">https://guzhaolin.art/blog/coding-agent-benchmark-boundary</guid><description>同一个模型换一套执行框架，Coding benchmark 为什么会明显上涨？本文拆开模型权重、训练轨迹、harness、推理预算和验证器各自贡献了什么。</description><pubDate>Fri, 26 Jun 2026 16:00:00 GMT</pubDate></item>
<item><title>AI 为什么永远还能再找出一个 P0？</title><link>https://guzhaolin.art/blog/ai-p0-review-loop</link><guid isPermaLink="true">https://guzhaolin.art/blog/ai-p0-review-loop</guid><description>微信支付的 90 天券和 30 天资金冻结期，暴露了 AI 为什么会漏掉业务后果，又为什么能在代码审查里无限制造 P0。</description><pubDate>Wed, 17 Jun 2026 16:00:00 GMT</pubDate></item>
<item><title>Agent 的“状态工程”到底在解决什么？</title><link>https://guzhaolin.art/blog/agent-state-engineering</link><guid isPermaLink="true">https://guzhaolin.art/blog/agent-state-engineering</guid><description>上下文窗口不能自动成为项目的权威状态。State Engineering 真正要解决的是：怎样重建此刻仍然有效的事实，并把当前决策需要的部分交给 Agent。</description><pubDate>Thu, 11 Jun 2026 16:00:00 GMT</pubDate></item>
<item><title>AI 产品价值判断与范式推演框架</title><link>https://guzhaolin.art/articles/ai-product.html</link><guid isPermaLink="true">https://guzhaolin.art/articles/ai-product.html</guid><description>一套用于判断 AI 产品价值、推演产品范式变化、识别真实机会与伪需求的系统化分析框架。</description><pubDate>Thu, 04 Jun 2026 16:00:00 GMT</pubDate></item>
<item><title>Agents 系统理解手册</title><link>https://guzhaolin.art/articles/books/agents-system-handbook/index.html</link><guid isPermaLink="true">https://guzhaolin.art/articles/books/agents-system-handbook/index.html</guid><description>一本围绕 Agent 概念、业务、模型、系统、协作、治理、研究、工程与框架层展开的方法论手册。</description><pubDate>Sat, 23 May 2026 16:00:00 GMT</pubDate></item>
<item><title>现代前后端完全指南：软件城市宇宙</title><link>https://guzhaolin.art/articles/books/software-city-guide/index.html</link><guid isPermaLink="true">https://guzhaolin.art/articles/books/software-city-guide/index.html</guid><description>用“软件城市”的方式组织现代前端、后端、数据、基础设施、AI 应用与工程化技术地图。</description><pubDate>Sat, 09 May 2026 16:00:00 GMT</pubDate></item>
</channel></rss>
