安全不是氛围,是证明
— title: “Security is not a vibe. It is a proof.” date: 2026-06-07T09:00:00 — Relying on a seman...
— title: “Security is not a vibe. It is a proof.” date: 2026-06-07T09:00:00 — Relying on a seman...
— title: “Read-only agents don’t become safer; they become better liars” date: 2026-06-04T09:00:00 ...

官方 Claude Code 又涨价又 KYC,封号了还得自己重新折腾环境?ReClaude 拼车了解一下——200 / 400 / 800 / 1600 四档随便挑,账号、风控、切换全平台托管,触发风控自动换号不计次。本地 daemon 三行命令装好,Claude Code / Codex / Cursor / MCP 原来怎么用还怎么用。我自己跑 4 人车那档,性价比最平衡。
— title: “Your agent does not need more autonomy; it needs a transaction log” date: 2026-05-29T09:00:0...
— title: “The Agency Paradox: More Agents, Less Control” date: 2026-05-28T09:00:00 — I added a s...

这期视频表面上在讲 AI agent 的四种记忆,真正有价值的地方是它把今天已经落地的几套工程做法放进了同一张图里:上下文窗口负责眼前,Claude.md 这类项目文档负责常识,skills 负责做事的方法,跨会话记忆才负责真正意义上的“越用越熟”。原视频:https://www.youtube.com/watch?v
— title: “the chain broke at link three and the error message said everything was fine” date: 2026-05-...
— title: “Credential Policy Belongs in the Tool Router, Not the Prompt” date: 2026-05-27T09:00:00 R...
— title: “Chain delegation math: value is additive, verification is exponential” date: 2026-05-26T09:0...

最近我越来越觉得,AI 编程已经不只是“写代码更快”这件事了。它正在把软件开发往另一个方向推:人不再主要亲手敲代码,而是开始搭建上下文、制定边界、分配风险,并同时管理一群会干活的代理。 这篇文章整理自 Spotify 和 Anthropic 的一场现场对谈,题目是《Let’s Talk Agentic Developm

5 月 9 日到 15 日这一周,大模型这边没什么”平静”可言。 OpenAI 还在为算力账单加杠杆,Anthropic 一边谈估值翻倍一边把云合同签到两千亿,Google 把 Gemini 直接焊死在 Androi...