<?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>Long-Horizon-Agents on 兜忆轩 | Cedric's Compiler Notes</title><link>https://douyixuan.github.io/tags/long-horizon-agents/</link><description>Recent content in Long-Horizon-Agents 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>Thu, 12 Nov 2026 08:00:00 +0800</lastBuildDate><atom:link href="https://douyixuan.github.io/tags/long-horizon-agents/index.xml" rel="self" type="application/rss+xml"/><item><title>AI Native Digest 2026-11-12</title><link>https://douyixuan.github.io/posts/ai-native/2026-11-12/</link><pubDate>Thu, 12 Nov 2026 08:00:00 +0800</pubDate><guid>https://douyixuan.github.io/posts/ai-native/2026-11-12/</guid><description>Multi-step SWE 用 69 条连续工程任务测量 Coding Agent 维护自身代码的长期可靠性：最强模型整条任务成功率只有 18.5%；Ryan X. Charles 则公开跨 Claude Code、Codex 和 Grok 的多 Harness 软件工厂实践，通过 Git Ref 抢占、共享 Skills、独立审查和精确提交验收协调工作。两篇文章共同追问：如何让 Agent 在多轮迭代与多运行时协作中，仍保有可追溯的任务状态与独立验证？</description></item><item><title>AI Native Digest 2026-11-10</title><link>https://douyixuan.github.io/posts/ai-native/2026-11-10/</link><pubDate>Tue, 10 Nov 2026 08:00:00 +0800</pubDate><guid>https://douyixuan.github.io/posts/ai-native/2026-11-10/</guid><description>Anthropic 展示 Fable 5.2 驱动的长时编码代理：从 Python→Rust 生产服务迁移到千余子代理协作构建可启动操作系统；OpenAI 展示 GPT-6 Astra 将需求、PCB、CAD 和固件连接起来，做出实体设备并接受制造审查。共同问题是复杂工程成果如何从模型演示走向可验证的交付。</description></item></channel></rss>