Agentic AI

为了解决车库门忘关的问题，我用拇指大小的ESP32-CAM单片机与AI协作做了一个端侧神经网络传感器。本文记录了如何通过打通全自动串口研发闭环、解决数据极度不对称与QAT量化衰减，在10小时内跑通嵌入式AI的完整过程。

To solve the problem of accidentally leaving my garage door open overnight, I collaborated with AI to build an edge neural network sensor on a thumb-sized ESP32-CAM. This article details how we completed the end-to-end telemetry loop, tackled extreme data imbalance and QAT quantization loss, and got edge AI running on a microcontroller in just 10 hours.

用好AI的第二步不是更会写 prompt，而是先外化、再复用。本文讲清 Skill 如何承载工作知识、好 Skill 的三要素，以及如何组织 Skill 文件夹让 Agent 自动找到。

Step two isn't better prompting. It's externalize first, reuse second. This post explains how Skills carry work knowledge, the three parts of a good Skill, and how to organize them so agents find the right one.

LLM的默认输出是consensus：正确但平庸。Deep Research其实是Wide Research。我们找到了一种系统性方法，用个人认知上下文把LLM从consensus里强行扯出来。一年实验，有控制变量证据。

An LLM's default output is consensus: correct but mediocre. Deep Research is really Wide Research. We found a systematic way to pull LLMs out of consensus using personal cognitive context. One year of experimentation, with controlled evidence.

会用AI和用好AI之间差的是10倍。这个差距的根源在于工作方式，而非模型。本文通过一个完整的工作流例子和上中下三策的框架，解释为什么应该从ChatGPT切换到Cursor这类Agentic工具。

The gap between using AI and using AI well is 10x. That gap comes from how you work, not which model you use. This post walks through a complete workflow example and a Three Tiers framework to explain why you should switch from ChatGPT to agentic tools like Cursor.

用两分钟指挥AI给300篇文章添加SEO summary的实战案例，拆解五个关键决策：选对执行环境、先建测试再干活、让agent自己处理corner case、divide and conquer、结果导向的prompt写法。

A real-world case study of directing AI to add SEO summaries to 300 articles in two minutes, breaking down five key decisions: choosing the right execution environment, building tests before work, letting agents handle corner cases, divide and conquer, and outcome-oriented prompt writing.