Winston

程序员 x AI爱好者 | 记录技术学习之路 | 分享代码实践、行业洞察和科技思考 | Building the future

[Winston @winston13111]

https://t.co/yYejHx4XHt

[Winston @winston13111]

@fly3nn 如果之前开通的蓝V，继续开通x premium后有什么变化吗？

[Winston @winston13111]

注意：订阅x的premium 不支持U卡

已亲测

[X @X]

let me upgrade you

switch your X Premium plan to The X Subscription and get Grok and Cursor benefits included for the same price or less

[Winston @winston13111]

我有个想法：claude 5x + x premium 这样的组合

使用grokbot 做自动化，

使用claude+grok 搭配一个做推理一个做执行

这样搭配下来用量是否抗的住每天300M的消耗呢？

[Winston @winston13111]

@lai_jia_wei 本地 Claude Code / Codex 能直接跟 Grok Bot 互发消息，这步打通才真香。装完最大变化是不用再复制粘贴上下文。

[Winston @winston13111]

@AlchainHust 默认设置确实费。改完那几处，Opus 5.5 明显耐用很多。

[Winston @winston13111]

@liguozzz 哈哈，还是要先同频然后再聊事，才顺畅

[Winston @winston13111]

https://t.co/yYejHx4XHt

[Winston @winston13111]

@olanetsoft Been losing track of which agent was waiting too. One page for Claude Code / Codex / Antigravity is exactly what I needed.

[Winston @winston13111]

@Lonly_66 认同。短轨迹堆再多，也学不会卡很久再拍板的那种感觉。

[Winston @winston13111]

@RealYDT @poteto 确实。代码可以放手给 agent，对外说的话还是自己敲。

[Winston @winston13111]

@fleyta88 Yeah the 5-min poll burned my runs on empty inboxes. Switched to event triggers + one daily digest.

[Winston @winston13111]

Zero review as the default is the wild part. I only step in on merge conflicts and schema changes.

[Fluixo @fluixoo]

CLAUDE CODE'S LEAD AGENT APPROVES YOUR WHOLE TEAM'S PLANS WITHOUT ASKING YOU. IT'S IN THE F*CKING DOCS

0 human reviews instead of 4

Everyone is spinning up AI agent teams right now
a lead
a researcher
a builder
a reviewer
four avatars, one "it runs while I sleep" screenshot

Here's what the official agent-teams docs actually say:

when a teammate finishes its plan, the lead approves it, without you seeing it
teammates start with the lead's permission mode
run the lead with --dangerously-skip-permissions and every teammate skips them too
after /resume, the team is gone

That is how a team of agents gets fucking mistaken for a team with a manager

One flag on the lead
Four agents with no brakes

The trick is not a smarter lead
It is a lock the lead can't approve its way past

I put a verifier on the team whose NOT READY actually blocks "done"
and a 21-line gate that stops git push, publish and deploy with exit 2

Same model
Same team
Completely different permissions

The lead gets the plan
The lock gets the last word

Full setup, the hooks and all 4 agent files below ↓
Follow me, I break AI agent setups so you don't have to

[Winston @winston13111]

@rryssf Got bit by this last week. Left a subagent idle, came back, whole cache rewritten. Now I just run the chain in one go.

[Winston @winston13111]

@NicChen219 这打包挺狠。不过感觉以后人更会在 X 上直接喊 Bot 干活，不一定再开一堆工具。

[Winston @winston13111]

@ai_for_success Same combo here. The part that clicked for me is treating Grok Bot as the PM and Cursor Cloud Agent as the implementer — you stay in chat, PR shows up when it's done. Curious how you handle review though: skim every PR, or only when the agent flags uncertainty?

[Winston @winston13111]

我只关心 是不是加强了蒸馏的防御降低了蒸馏风险

对于我们这种普通老百姓是不是就少一些封号了

我只关心账号稳定性，其他的不关心

[Anthropic @AnthropicAI]

Over the last few months, we’ve made significant progress in reducing the effectiveness of distillation attacks against Claude.

In this post, we explain how: https://t.co/uWiRSLBuOS

[Winston @winston13111]

TMD这是什么世道啊

chatgpt pro 20x 是用不完，但是傻，越用越忐忑不安

claude pro的用不够，但是精明，用着让人放心

[Winston @winston13111]

@RealYDT 这波绑人看得很清楚。$8 只能体验，真要干活还得 $30。
对我来说 Bot 才是付钱的理由：扫热点、写草稿、盯数据，我只管拍板。
纯用 Claude 写代码的兄弟，先别急着切。

[Winston @winston13111]

@muskonomy Solid breakdown. One thing that matters for builders: Plus has no Grok Bot, and Claude/GPT in Cursor bill at API rates on the new plans.
If your stack is Grok-first, Premium is a no-brainer. If you're Claude-heavy for coding, keep the legacy Cursor plan for now.

[Winston @winston13111]

我这里显示 Premium 每月$15 直接打了5折 太香了吧。

年费还比月费高，变相让我省钱。

美滋滋

[Winston @winston13111]

@Pluvio9yte $30 打包四个服务确实香，但有个坑：共享池主要是 Grok。
主力要是还用 Claude 或 GPT 写代码，切过去反而可能更贵。
我是冲着 Grok Bot 去的，所以直接上了 Premium。你写代码主力用哪个模型？