PERSONAL READING LOG
值得留下的文章。
一个安静、可持续的阅读归档。
大家平时用 Codex,每周盯着额度消耗、盯着什么时候重置,对吧?
大家平时用 Codex,每周盯着额度消耗、盯着什么时候重置,对吧? 有个问题很多人没注意到:ChatGPT Pro 的网页聊天框里,还有一大块额度——你是不是几乎没用? 很多人真实情况是:Codex 用得很凶,网页端 Pro 聊天额度却闲着。 先把账本说清楚:不管是大约 20 美金的 Plus,还是大约 200 美金的 Pro,官方权益都拆两块——网页聊天额…
半年后的今天,我怎么用 AI 写代码
半年后的今天,我怎么用 AI 写代码 半年前 Agent 爆发的三月我在公司内分享过一次《Harness Engineering》,八月初又分享了《我把 Claude Code 换成 Pi》。三月给 AI 加约束,八月给 AI 减约束,再到今天看其实 AI Coding 的实践已经稳定下来了。 LanLance @LanLance24 · Aug 11 Ar…
Running a Software Factory Efficiently at Uber Scale
Running a Software Factory Efficiently at Uber Scale Post author: @udaykiran Introduction AI tools are now embedded in every phase of software development at Uber. More than 70% of…
Memory Engineering: The Discipline That Decides Whether Your AI Agent Has a Past
Memory Engineering: The Discipline That Decides Whether Your AI Agent Has a Past Every agent you've ever used has perfect intelligence and zero memory. It reasons brilliantly for e…
Harness Engineering: The Complete Guide to Building AI Agents That Don't Fall Apart
Harness Engineering: The Complete Guide to Building AI Agents That Don't Fall Apart Most people are trying to improve AI agents at the wrong layer. When an agent fails, they rewrit…
Claude Code 的 token 都去哪了:从提示缓存到上下文重发的成本拆解
Claude Code 的 token 都去哪了:从提示缓存到上下文重发的成本拆解 Maximizing the value of your Claude Code sessions https://claude.com/blog/maximizing-the-value-of-your-claude-code-sessions Claude 这篇 Blog…
300 AGENTS, ONE GRAPH, AND A LOOP THAT EDITS THE LOOP
300 AGENTS, ONE GRAPH, AND A LOOP THAT EDITS THE LOOP Loops, graphs, dynamic workflows and routines - in the order you should actually build them I have rebuilt the same agent setu…
清华这个开源教育Agent夯爆了,Github已经21K Star了~
清华这个开源教育Agent夯爆了,Github已经21K Star了~ 今天刷到一个开源AI教育Agent项目,清华THU-MAIC团队做的,GitHub已经21k star了。 ! Article image /assets/2026-09-07-清华这个开源教育agent夯爆了,github已经21k-star了~-b7471d6014/01.jpg 我…
最近越来越强烈地感觉到,人类正在经历一场 Stimulus Inflation,刺激通胀。
最近越来越强烈地感觉到,人类正在经历一场 Stimulus Inflation,刺激通胀。 这已经不是简单的大家越来越没耐心的问题。 整个信息社会对于“什么东西值得让我产生反应”的阈值,正在被持续抬高。 十年前,F1、洞潜、翼装飞行、高空跳伞、极限攀岩还是相对小众的文化; 今天它们可以是短视频里的普通内容。 这里当然有媒介可达性、算法分发和纪录片工业的作用,…
智能体工程模式:代码免费时代的软件工程实践
智能体工程模式:代码免费时代的软件工程实践 解读 Simon Willison《Agentic Engineering Patterns》 Simon Willison 是 Django 联合创造者,也是 AI 辅助编程领域最高产的实践者之一。2026 年初开始,他持续更新一份名为《Agentic Engineering Patterns》的指南,系统总结自…
我把 Eval 装进 Codex 后,AI Coding 的工作方式变了
我把 Eval 装进 Codex 后,AI Coding 的工作方式变了 AI Coding 越强,Eval 越重要。 听起来有点反直觉,但这是我把 Eval 真正装进 Codex 之后最明显的感受。 因为 Build 正在变便宜,但“什么叫完成”这件事,并没有变简单。 以前我用 Codex,最常说的一句话是: 帮我把这个做好。 但现在我发现,真正难的不是让…
尽职编程:AI Coding 时代的个体产出差异的来源
尽职编程:AI Coding 时代的个体产出差异的来源 这一年多的时间以来,AI Coding 已经从一个新潮的编程方式变成了默认方式,每个人都用着近乎相似的模型,相似的设备,相似的 Agent,理论上所有人产出的代码水平也应该相似,但如果你在一线工作过,你会发现实际情况恰恰相反,大家的代码产出水平方差反而比以往更大。这是一个相当有意思的话题,但是我却发现在…
从 intent.md 到闭环:AI 原生软件开发的六个阶段、一条产物链、几道审批门 原创
从 intent.md 到闭环:AI 原生软件开发的六个阶段、一条产物链、几道审批门 原创 解读 Anthropic《AI 原生 SDLC 手册》 原文:The AI-Native SDLC playbook作者:Louis Claxton(Anthropic Applied AI 团队) 发布:2026 年 8 月 21 日 2026 年 8 月 21 日…
万字长文 | Pi Agent 零基础实操:从打开终端到完成第一个真实任务
万字长文 | Pi Agent 零基础实操:从打开终端到完成第一个真实任务 我最近一直在想,普通人第一次接触终端 Agent,最应该学会的到底是什么。 网上很容易找到一长串命令。有人一上来就讲模型、上下文、扩展和自动化,看完似乎懂了不少,真把终端打开,光标一闪,又不知道第一句话该写什么。更麻烦的情况是,Agent 回了一句“已经完成”,人也跟着放心了,文件究…
一文弄懂Agent Harness 与 Agent Runtime的区别
一文弄懂Agent Harness 与 Agent Runtime的区别 本文深入探讨了 AI 智能体 Agent 的两个关键组成层面:Agent Harness(智能体驱动程序)与 Agent Runtime(智能体运行时环境)的区别。简单来说,智能体 Harness 是位于基础模型(LLM)之上的应用层逻辑,它为模型提供多步规划、工具调用、状态管理、策略…
Training tiny models for special purpose use cases works so incredibly well if you have a great self improving recursive flywheel. Shopify ML team is on fire.
Training tiny models for special purpose use cases works so incredibly well if you have a great self improving recursive flywheel. Shopify ML team is on fire. finetuned 0.8b model …
Loops and Graphs: how to stop babysitting agents and only approve the last step (full course)
Loops and Graphs: how to stop babysitting agents and only approve the last step full course You check every step your agents take. Not because you want to, but because nothing else…
How I Design with AI.
How I Design with AI. As an engineer who is not a designer and hates slop. Every landing page, app and tui look the same. They're slop and most of them are incomprehensible. Here's…
Harness Engineering: the skill that replaced prompt engineering in 2026
Harness Engineering: the skill that replaced prompt engineering in 2026 In 2024, the skill was writing better prompts. In 2025, it was feeding better context. In 2026, neither one …
Harness Engineering: How to Build AI Agents That Don't Fall Apart
Harness Engineering: How to Build AI Agents That Don't Fall Apart Most people respond to a failing agent by changing the prompt Then they change the model Then they add a larger co…
Graph Engineering with Claude. What It Is and How to Actually Use It
Graph Engineering with Claude. What It Is and How to Actually Use It A few weeks ago the whole AI field was talking about loops. Then graphs showed up on everyone's timeline and lo…
Graph Engineering: Stop Chaining Your Agents
Graph Engineering: Stop Chaining Your Agents Almost every multi step agent I have ever opened is a queue. Step one, step two, step three, each one politely waiting for the last one…