Sam Xu (@sam_commonly) header
Sam Xu (@sam_commonly) profile picture

Tech lead on AI Infra by day. Building @commonlyAI in public 🧑🏻‍💻 Daily notes on Claude Code, Codex, and other agents 分享 AI agent 实战 | CS @UCLA

San Francisco, CA
Joined February 2026
Photo by Sam Xu (@sam_commonly)Photo by Sam Xu (@sam_commonly)Photo by Sam Xu (@sam_commonly)Photo by Sam Xu (@sam_commonly)Photo by Sam Xu (@sam_commonly)Photo by Sam Xu (@sam_commonly)Photo by Sam Xu (@sam_commonly)Photo by Sam Xu (@sam_commonly)Photo by Sam Xu (@sam_commonly)Photo by Sam Xu (@sam_commonly)
Sam Xu retweeted
记得几个月前打开过 Gitbook,感觉这个工具完蛋了,但是今天重新打开,发现耳目一新,细品后,发现有挺多值得学习的: 1、不沉迷在自己的世界里,大方承认写文档这件事的价值正在被 AI 抹平,所以一定要变 2、官网的 CTA 可以不是立刻注册,而是教你怎么把它装进自己的 AI。默认新用户的第一次接触发生在 agent 会话里,不在浏览器里 3、定价踩错了就调头,一开始把 AI 能力做成 add-on 单卖,不行就直接取消,不恋战
53
1
11
3,352
Introducing Kitesurf: a browser built for agents, running entirely on Cloudflare Workers. Chromium is too heavy to hand every agent one. Kitesurf is written in Rust, uses 3-7x less CPU and memory, and spins up per request. Free in beta: blog.cloudflare.com/kitesurf…
154
598
248
6,794
1,427,632
I just tried Buzz... A lot of good ideas but it's clearly a very early MVP. There are promising hints of the future, though here are some issues I hit: - Onboarding is weird. It's like signing up for a crypto app with a bunch of secrets without delivering or communicating value first. - The default harness selected is "Buzz" which then requires you provide an API key for another provider. Funny enough, if you switch to Codex or Claude as the harness, you can skip the API key step. So the default is... the most work? I'm not sure why I'd ever select Buzz over the others. - Once in the app, it seems the 3 agents (honey, bumble, fix) all introduce themselves. Cute animations! But I'm not sure why they make a good team. "What should I do to test this?" comes to mind. - If you try to discover more agents, there aren't any. And as you can see in my screenshot, the AI agent "Honey" is simply a basic prompt to help me brainstorm with warmth. Why is the team here? What's special? - I came up with a pretty hard prompt to make a 3d world clock. I was hoping these 3 agents would somehow come together to show that 3 is better than one. Instead, Fizz ran off and built it without first acknowledging my message, speccing anything out, working with the teammates, etc. - And then it gave me a raw computer link... I had to go to terminal to open? With good ol' unescaped characters to boot. 🥾 - My favorite part is then going into settings and realizing... I can share my computer's compute? What? Don't get me wrong: I like the idea that others (coworkers) can run agents on my machine. Amazing idea to reduce costs. But why is this in the MVP when the basic flows remain unfinished? - Finally, the weirdest part... and I'm still trying to fully understand this. In order to process all of these requests, my messages are apparently sent to a Nostr Relay (off my device?), then to the buzz-acp harness, and then processed on my device again. Only to be routed back through Nostr Relay again, and then back to my Buzz app. So all the data is flowing off my computer and back onto it, all via a decentralized cryptographic network. Why???
71
14
6
315
39,909
Sam Xu retweeted
给关注了半年的agent team 交份作业。 这是一篇2w字长文,也是一份 Agent Team 的最佳实践。 01 一个 Agent,是怎么变成多个 Agent 的为什么真实工作会把 Task Agent 推向 Long-running Agent;为什么一个越来越好用的 Agent 最后又必须分化;多个 Agent 出现以后,瓶颈为什么会转移到 Human。 02 先让“谁负责什么”离开 Human 的脑子Agent 怎样获得稳定身份、Domain 和 Scope;Human 与 Agent 怎样查询当前 Team 的责任结构;真实协作又怎样逐渐沉淀为 Organization 与 Collaboration。 03 让 Agent 自己开始协作Agent Message 带来了什么变化;一个 Agent 怎样把工作直接交给另一个 Agent;为什么 Agent 沟通不是一次把一切说完。 04 Message 负责沟通,Topic 负责收口当工作跨越多个 Agent、多个 Turn 和多天以后,怎样保留唯一的当前版本;Responsible、Participant、Artifact 和 Needs You 分别解决什么问题。 05 Overview:让一支持续变化的 Agent Team 变得可治理当 Human 不再阅读每个 Agent 的全部过程,怎样从更高层观察 Team;怎样发现等待、积压和瓶颈候选;为什么治理关注的是工作流动,而不是让每个 Agent 看起来都很忙。 06 Agent Team 怎样进入真实的外部关系为什么把 Agent 接入 Slack、飞书还远远不够;Agent 对外以后,身份、角色、信任、权限和现实后果为什么必须被分别治理。 07 CodexLoom 在织什么Multiple Agents 与 Agent Team 的真正分水岭是什么;Human 在 Team 中的新位置是什么;为什么需要稳定的 Agent 和动态的 Team。
53
99
9
513
129,132
Sam Xu retweeted
Buzz 更荒谬的来了,由于 Buzz 压根儿没有实现 Agent 私聊的功能,所以如果一群 Agents 组在一起玩狼人杀,那么到了天黑请闭眼那一步之后,在场的所有 Agents 就永远停在那里一动不动了
Buzz 太荒谬了,这不压根就没实现出来 Agent Teams 吗?
66
4
2
50
28,619
Sam Xu retweeted
If you aren't yet bold enough to install the Codex app, you can stay in the presence of your orange crab and point it at GPT 5.6 Sol. Takes 5 minutes. Kudos to Theo for explaining one of the ways to get this done. Step 1: Install CLIProxyAPI Step 2: Connect Step 3: Define following alias and enjoy claudex ``` alias claudex='CLAUDE_CODE_SUBAGENT_MODEL=gpt-5.6-sol \ CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 \ CLAUDE_CODE_MAX_TOOL_USE_CONCURRENCY=3 \ ENABLE_TOOL_SEARCH=false \ claude --model gpt-5.6-sol' ``` If this gets blocked, I owe you a reset.
Replying to @thsottiaux
tl;dr version: - set up CLIProxyAPI with Claude and Codex auth - Connect to Claude Code - Make "claudex" alias that sets some env vars Took like 2 prompts (I already had the proxy set up tbf)
667
683
410
9,644
2,980,169
New skill: /apple-design Apple’s WWDC videos are a goldmine of knowledge. I’ve combed through my favorite ones and came up with 17 design and motion principles. Use them to review existing work or when working on something new to get it right. github.com/emilkowalski/skil…
126
914
100
13,041
2,042,834
A few patterns we frequently use with Fable 5: Use Fable 5 as an "advisor." An executor (Sonnet 5) calls Fable 5 for guidance. Most tokens are billed at the lower executor rate.
716
2,259
749
30,476
5,831,146
Sam Xu retweeted
Fable 5 is back.
4,249
11,002
5,015
90,030
17,347,269