chynnahe

独立开发者｜做产品，也做增长｜记录从0到1的实践与复盘

[chynnahe @chynnahe1]

第一次见有人在开着codex pro套餐的同时，还在kimi花了7000多，据说是一直在买加油包，kimi这么贵吗

[chynnahe @chynnahe1]

@peter45174526 相当牛了

[chynnahe @chynnahe1]

@jiubencong 老哥厉害

[chynnahe @chynnahe1]

刚看了眼 X 的奖励门槛：90 天 50 万首页曝光。
我现在 95。
行吧，先假装没看见，继续发。

[chynnahe @chynnahe1]

@chngyqng787347 一起加油啊

[chynnahe @chynnahe1]

刚说曝光不行呢，这就好起来了吗，哈哈

[chynnahe @chynnahe1]

大A今天又是因为啥，有没有什么是我不知道的

[chynnahe @chynnahe1]

@xiax0603 回访这么有用啊

[chynnahe @chynnahe1]

现在打开 X ，像误入创作者绩效会：
有人一个月涨 5.5 万粉，
有人三天只有 28 个曝光，
还有人昨天手搓了 658 条回复。

我的500关注已经到了，啥时候曝光能达到啊，已经开始紧张了。

[chynnahe @chynnahe1]

这是什么错误，不让我回关了

[chynnahe @chynnahe1]

挺快，破500了

继续交朋友

[chynnahe @chynnahe1]

@Deepinstorm 来了

[chynnahe @chynnahe1]

各位早上说好啊，又开始浇朋友了

[chynnahe @chynnahe1]

这广告创业也是绝了，一定要看到最后

[chynnahe @chynnahe1]

这条里我最看重的不是“哪个模型做哪一步”，而是后面的决策记录、硬规则和停止检查。

Agent 真正进入工作流后，关键是：出错时能否追溯依据，遇到高风险动作能否停下，执行失败能否回退。

成本数字可以另行验证；这些边界才决定它能不能放心交活。

[Mr. Buzzoni @polydao]

I STOPPED LETTING CLAUDE MAKE DECISIONS THE DAY I BUILT THIS JEV AGENT FOLDER

I used to let Claude decide, write and act on every single step

-> now Claude only writes. Jev makes the calls, code does the acting, and every step leaves a receipt

here's what's inside the folder:

• the input
> AGENTS.md - when to call Jev, when to skip it
> state/build_state.py - goal, workers, done, missing, constraint. evidence, never a summary

• Jev decides (questions/)
> https://t.co/SIhqwoheZP - which model tier gets the task
> next_worker.py - which worker acts next
> https://t.co/y6q9a7fv9Y - keep or drop each tool output
> https://t.co/8LKHJE7ySk - is the goal actually met
> https://t.co/WtQhAqPFSG - does this send, pay or delete?

• code acts (rules/)
> hard_rules.py - stop after ten actions, never publish unapproved
> thresholds.yaml - act only on confident answers. fraud needs 0.95

• Claude writes (workers/)
> https://t.co/Pl96LqadxQ - sources and notes
> https://t.co/4GCndJWKsk - drafts and briefings
> the only place in the whole folder where text gets generated

• the proof
> receipts/decisions.jsonl - options offered, chosen id, re-check, fallback
> evals/ - dozens of my own labelled traces, thresholds tuned on them

• the guards (hooks/)
> pre_tool_use.sh - every command checked before it runs
> https://t.co/UC0IBplGsa - checks "all done" before the agent is allowed to stop

10,000 decisions for $0.42. median 300 ms

an LLM writes, Jev decides, code acts

[chynnahe @chynnahe1]

https://t.co/NNiQVR6EZS

[chynnahe @chynnahe1]

@xiax0603 没记错的话下周还有两张重置卡到期

[chynnahe @chynnahe1]

听说codex出500刀的订阅了？

[chynnahe @chynnahe1]

微软公布了 Copilot Autopilot，连催进度都给安排上了。
给它一个角色和目标，它可以盯 Teams 频道、跟进讨论、主动找同事问进展。重复任务定期做，项目隔几天还能接着跟，你离线后它也继续跑。
它有自己的记忆和工作空间。微软举的例子是供应商评审：排计划、准备会议、跟后续进展，都交给它推进。
9 月底扩大私有预览，还没全面开放。
https://t.co/KJZ4hNJ8jA

[chynnahe @chynnahe1]

@OrientLinden 感谢老板