<?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>Agent-Infrastructure on 兜忆轩 | Cedric's Compiler Notes</title><link>https://douyixuan.github.io/tags/agent-infrastructure/</link><description>Recent content in Agent-Infrastructure 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>Mon, 21 Sep 2026 08:00:00 +0800</lastBuildDate><atom:link href="https://douyixuan.github.io/tags/agent-infrastructure/index.xml" rel="self" type="application/rss+xml"/><item><title>AI Native Digest 2026-09-21</title><link>https://douyixuan.github.io/posts/ai-native/2026-09-21/</link><pubDate>Mon, 21 Sep 2026 08:00:00 +0800</pubDate><guid>https://douyixuan.github.io/posts/ai-native/2026-09-21/</guid><description>今天关注 AI-native 工程的两个底层问题：Enola 解释为什么真正决定 coding agent 能力边界的是 harness 对 context、tools、persistent state 和 feedback 的选择；ARIMLABS 则展示了 agent eval 从本地实验扩展到每天约 50 万次执行时，isolated computers、parallel trials、network policy 与 inspectable reruns 如何成为评测基础设施。</description></item></channel></rss>