<?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>技术深潜 on Symbol's Notes</title><link>https://hongsymbol.online/categories/%E6%8A%80%E6%9C%AF%E6%B7%B1%E6%BD%9C/</link><description>Recent content in 技术深潜 on Symbol's Notes</description><generator>Hugo</generator><language>zh-cn</language><lastBuildDate>Mon, 11 May 2026 20:00:00 +0800</lastBuildDate><atom:link href="https://hongsymbol.online/categories/%E6%8A%80%E6%9C%AF%E6%B7%B1%E6%BD%9C/index.xml" rel="self" type="application/rss+xml"/><item><title>Harness Engineering：让 AI Agent 可靠工作的完整方法论</title><link>https://hongsymbol.online/posts/harness-engineering/</link><pubDate>Mon, 11 May 2026 20:00:00 +0800</pubDate><guid>https://hongsymbol.online/posts/harness-engineering/</guid><description>Agent 效果不好，不一定是模型的问题，很可能是你的 Harness 不够好。读完《Learn Harness Engineering》127页后的精华提炼——五子系统、12讲核心要点、可立即行动的实践清单。</description></item><item><title>假如 LLM 无限上下文了，RAG 还有意义吗？</title><link>https://hongsymbol.online/posts/rag-vs-infinite-context/</link><pubDate>Mon, 11 May 2026 19:30:00 +0800</pubDate><guid>https://hongsymbol.online/posts/rag-vs-infinite-context/</guid><description>上下文窗口从 4K 飙到 200 万 token，很多人觉得 RAG 要凉了。但这个判断错在三个层面。RAG 不会消亡，而是在进化——从 Naive RAG 到 Agentic RAG，解析 5 个 RAG 依然不可替代的理由。</description></item><item><title>Agent 评测到底怎么做？告别「感觉它还行」</title><link>https://hongsymbol.online/posts/agent-eval/</link><pubDate>Mon, 11 May 2026 19:00:00 +0800</pubDate><guid>https://hongsymbol.online/posts/agent-eval/</guid><description>改了 Prompt 感觉没问题→上线→用户反馈变蠢了→永远在救火。从 Vibe Check 进化到 EDD（评测驱动开发），Anthropic 工程博客精华提炼——三种评分器、两套评测套件、pass@k vs pass^k 核心指标，8步从零搭建评测体系。</description></item><item><title>从零构建 AI Agent：架构设计与实践总结</title><link>https://hongsymbol.online/posts/building-ai-agent/</link><pubDate>Sun, 15 Mar 2026 11:00:00 +0800</pubDate><guid>https://hongsymbol.online/posts/building-ai-agent/</guid><description>基于在百度贴吧 Agent 社区的实战经验，总结 AI Agent 从架构设计到上线运营的核心思考。</description></item><item><title>Prompt Engineering 实战指南：从 Zero-Shot 到 Multi-Agent</title><link>https://hongsymbol.online/posts/prompt-engineering-guide/</link><pubDate>Sat, 28 Feb 2026 14:00:00 +0800</pubDate><guid>https://hongsymbol.online/posts/prompt-engineering-guide/</guid><description>一份来自实战的 Prompt Engineering 指南，涵盖 Zero-Shot、Few-Shot、CoT、ReAct 等核心 pattern，附带真实业务案例。</description></item></channel></rss>