<?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/"><channel><title>Self-Improving-Agent on 兜忆轩 | Cedric's Compiler Notes</title><link>https://douyixuan.github.io/tags/self-improving-agent/</link><description>Recent content in Self-Improving-Agent on 兜忆轩 | Cedric's Compiler Notes</description><image><title>兜忆轩 | Cedric's Compiler Notes</title><url>https://douyixuan.github.io/images/papermod-cover.png</url><link>https://douyixuan.github.io/images/papermod-cover.png</link></image><generator>Hugo -- 0.146.0</generator><language>en-us</language><lastBuildDate>Sun, 30 Aug 2026 08:00:00 +0800</lastBuildDate><atom:link href="https://douyixuan.github.io/tags/self-improving-agent/index.xml" rel="self" type="application/rss+xml"/><item><title>AI Native Digest 2026-08-30</title><link>https://douyixuan.github.io/posts/ai-native/2026-08-30/</link><pubDate>Sun, 30 Aug 2026 08:00:00 +0800</pubDate><guid>https://douyixuan.github.io/posts/ai-native/2026-08-30/</guid><description>今天两篇分别看 AI-native 软件工程的两个关键层：GitLab 讨论代码生成趋于廉价后，软件平台的稀缺资源如何从 implementation 转向 context、verification、governance 和 evidence；Google Research 的 WikiSkill 则把 Agent 经验拆成 raw traces、persistent wiki 和 executable skills，让失败经验也能持续沉淀并推动 skill evolution。</description></item><item><title>AI Native Digest 2026-08-29</title><link>https://douyixuan.github.io/posts/ai-native/2026-08-29/</link><pubDate>Sat, 29 Aug 2026 08:00:00 +0800</pubDate><guid>https://douyixuan.github.io/posts/ai-native/2026-08-29/</guid><description>今天两篇都在回答 Agent 如何持续变强：JIT-Agent 把 harness 从人工维护的固定脚手架变成按任务动态生成、可修复、可演化的机器生成程序；Warp 则给出一个非常简单的生产实践，用 base skill + improver skill + human feedback，把 code review、spec 和 issue triage 的反馈持续折叠回可版本化的 Agent Skills。</description></item><item><title>AI Native Digest 2026-08-20</title><link>https://douyixuan.github.io/posts/ai-native/2026-08-20/</link><pubDate>Thu, 20 Aug 2026 08:00:00 +0800</pubDate><guid>https://douyixuan.github.io/posts/ai-native/2026-08-20/</guid><description>今天关注两个 AI-native 系统层问题：如何把知识工作 Agent 的搜索、编辑、review 和交付绑定到同一版本化 workspace，以及为什么 self-improving agent 的 memory loop 会放大随机性并依赖隐藏的任务顺序。</description></item></channel></rss>