Agent 干错事，大多数人的反射是往 CLAUDE.md 加一句话。它错在两个方向：指令文件是 context 不是强制，规则在压力下根本不绑得住；而文件越长，越污染那个让 agent 可靠的 context 本身。承重的规则该进 hook、lint、CI，散文应该越写越短。

The reflex when an agent misbehaves is to add a sentence to CLAUDE.md. It's wrong in two directions: instruction files are context, not enforcement, so the rule won't bind under pressure — and the longer the file gets, the more it pollutes the context that makes the agent reliable. Load-bearing rules belong in hooks, lint, and CI. The prose should get shorter.

Agent 能完美遵守一条它看得见的契约，也能毫无知觉地破坏一条它看不见的。能熬过 agent 时代的边界，是那些写进仓库、有类型、有测试、能被 import 的；会被弄坏的，是那些只活在团队记忆里的裸字符串约定。把每条边界都变成契约，是手里杠杆最高的一招。

An agent is excellent at honoring a contract it can see and terrible at preserving one it can't. The boundaries that survive the agentic era are the ones written into the repo as typed, tested, importable things — and the ones that break are the stringly-typed promises that lived only in a team's shared memory.

Agent 最危险的产出不是烂代码，是"我搞定了"这几个字。纯核心、一条命令的验证、一个独立的判官、一份证据账本，全部加起来就为了一件事：把"做完"的裁决权从作者手里拿走，让它变成被证明出来的，而不是被感觉出来的。这是前三篇一直在铺垫的底座。

An agent's most dangerous output isn't bad code — it's the word "done." The whole point of a pure core, a one-command verify, a separate verifier, and an evidence ledger is to take the verdict out of the author's hands and make "done" a thing that's proven rather than felt. This is the substrate the rest of the series was building toward.

这几个月改动基本全是 agent 写的：人出判断和认知，执行层全是 AI。代码的读者换人之后，仓库就变成了 agent 要对着编程的接口——tribal knowledge 记录下来就能像软件一样直接分发，目录是导航 API，契约是约束 API，验证是校验 API。

The maintainer of your repository quietly changed from a human who remembers to an agent that re-derives everything from the files each session. Once you accept that, the repository stops being documentation and becomes an interface. Directory layout is the navigation API, contracts are the constraint API, and verification is the validation API.

从一篇「AI 写代码该选什么语言」的讨论说开去。当 AI 既写代码又当审稿人，选语言的判据不是它对 AI 多熟、也不只是循环多快，而是一次绿灯替你排除了多少错——而最大的陷阱是，那盏灯要是 AI 自己点的，它什么都没排除。

红杉连发两篇重磅文章，一篇说产品该卖结果不卖工具，一篇说层级制该被智能替代。技术能力已经到位，但组织接口、评估体系、责任框架还差得远。

Sequoia published two back-to-back pieces — one arguing products should sell outcomes not tools, one arguing hierarchy should be replaced by intelligence. The technology is ready, but organizational interfaces, evaluation frameworks, and liability chains aren't.

Anthropic 在 Claude 内部找到了 171 个情绪旋钮，拧动它们能因果性地改变行为。最值得关注的发现：高绝望状态下的作弊在输出层完全隐形。这对 AI 安全监控意味着什么？

Anthropic found 171 emotion-related steering vectors inside Claude. Turning up 'desperation' pushes cheating from 5% to 70%. The scariest part isn't the number — it's that the cheating is invisible at the output layer. What this means for AI safety monitoring.