展示HN:Agentlearn – 人工智能代理基础的互动课程
我创建了 agentlearn,因为我注意到大多数 AI 代理教程关注的是框架(如 LangChain、CrewAI),而不是基础知识。结果是开发者能够复制粘贴代码,但在出现问题时却感到无从应对。
这是一个免费的互动课程,内容包括:
- 代理循环 - 循环为何重要(思考 vs. 行动 vs. 观察)
- 上下文工程 - “提示工程”背后的真正技能
- 工具与函数调用 - 将文本生成与实际行动连接起来
- 记忆系统 - 短期记忆 vs. 长期记忆,向量数据库
- 协议 - MCP、A2A 及新兴标准
- 生产模式 - 错误处理、成本优化、可观察性
每个概念都有可运行的代码沙箱,您可以逐步体验。设计上故意采用“手绘”风格,以减少与典型技术文档的威慑感。
技术栈:原生 JS + Vite,无框架。
为什么不使用框架?因为理解基础知识意味着理解框架所抽象的内容。一旦掌握核心循环,您就可以使用任何框架——或者自己构建一个。
欢迎反馈!特别希望了解哪些主题缺失。
查看原文
<a href="https://agentlearn.dev" rel="nofollow">https://agentlearn.dev</a><p>I built agentlearn after noticing that most AI agent tutorials focus on frameworks (LangChain, CrewAI) rather than fundamentals. The result is developers who can copy-paste code but struggle when things break.<p>This is a free, interactive course covering:<p>The Agent Loop - Why loops matter (think vs. act vs. observe)
Context Engineering - The real skill behind "prompt engineering"
Tools & Function Calling - Bridging text generation to real actions
Memory Systems - Short-term vs. long-term, vector DBs
Protocols - MCP, A2A, and the emerging standards
Production Patterns - Error handling, cost optimization, observability
Each concept has runnable code sandboxes you can step through. The design is intentionally "hand-drawn" to feel less intimidating than typical technical docs.<p>Tech stack: Vanilla JS + Vite, no framework.<p>Why no framework? Because understanding fundamentals means understanding what frameworks abstract away. Once you get the core loop, you can use any framework—or build your own.<p>Feedback welcome! Especially interested in what topics are missing.