# Claude Code 發布週年演進為自主 Agent 生態

> 📖 本站完整內容索引（documentation index）：[llms.txt](/llms.txt)

> 原作者：Boris Cherny (@bcherny) · 策展與摘要：EasyVibeCoding · 平台：X (Twitter) · 熱度：🔥 · 日期：2026-06-09

> 原始來源：https://x.com/bcherny/status/2064034799711588805

## 中文摘要

Claude Code 發布週年演進為自主 Agent 生態。

**核心轉變與自動化**
Boris Cherny 指出，Claude Code 的開發模式已發生根本性變化，重點如下：
- **自動模式（Auto Mode）取代計劃模式**：隨著模型能力提升，已不再需要手動規劃步驟，自動模式能更高效地處理任務，並透過內建的安全性分類器確保執行過程的可靠性。
- **例程（Routines）與循環（Loops）**：透過設定例程，Agent 可自動監聽 GitHub Issue、錯誤報告並主動提出修復方案，甚至在開發者介入前就完成 PR 提交與驗證。
- **上下文極簡主義**：不再過度依賴提示工程，而是提供必要的工具與方法，讓模型自主提取上下文，避免微管理導致的效率下降。

**工程組織的典範轉移**
Boris Cherny 強調，Claude Code 的導入改變了 Anthropic 的工作流程，使 AI 成為企業運作的核心：
- **角色融合**：工程師、設計師、PM 與財務團隊皆能透過 Claude Code 直接參與程式開發與資料分析，打破了傳統職能分工。
- **端到端所有權**：開發者不再受限於繁瑣的程式碼審查與 CI 維護，能更專注於產品構思與使用者互動。
- **行動化開發**：透過遠端控制技術，Boris Cherny 現在能直接在行動裝置上監控並啟動 Agent 執行任務，實現了隨時隨地的開發體驗。

<video src="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1781010676218-j3jj4ook.mp4" poster="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1781010702678-htLaC2j8hqdefaultjpg.jpg" controls playsinline preload="metadata" style="max-width:100%;height:auto;display:block;margin:1rem 0"></video>
> Reflecting on a year of Claude Code

**對未來的展望**
影片中提到，Agent 的未來將朝向更長運行時間與更高自主性發展。Boris Cherny 認為，過去幾年累積的傳統工程經驗在 AI 時代需要重新學習，團隊應保持好奇心，透過與使用者共同構建，探索 Agent 在不同場景下的新應用形式。

## 媒體內容

**Reflecting on a year of Claude Code**

**逐字稿**

- `00:00` 當我們剛發布 Claude Code 時，那就像是一支小影片，我記得我把它貼到 Slack 上，（When we first released Claude Code, it was like a little video, and I remember posting it to Slack,）
- `00:04` 當時大概只有兩個人給了反應，大家看起來都很興奮。（and there was like two people that gave like the reaction, and like people were like excited.）
- `00:09` 沒錯，我覺得這真的很酷，特別是對於我那些非常簡單的工程任務來說。（Yeah, I thought it was really cool, especially for my very easy engineering tasks.）
- `00:13` 它處理得相當不錯。（It was quite good at it.）
- `00:14` 這真是一個很好的說法，委婉地表示它其實沒那麼好。（That's like a really nice way to say that it wasn't really good.）
- `00:30` 真不敢相信我們發布 Claude Code 才過了一年。（I can't believe it's only been a year since we first launched Claude Code.）
- `00:34` 很難想像當時的情況是什麼樣子。（It's hard to remember what that was like.）
- `00:37` 那與我們今天所做的相比，簡直天差地遠。（It's so different than what we're doing today.）
- `00:40` 現在我手上有成群的 Agent 在執行各種任務。（Now I just have armies of agents that are doing stuff.）
- `00:44` 我正在對一個 Agent 下 prompt，或者我有一個 Agent 在對其他 Agent 下 prompt，而那些 Agent 又在對更多的 Agent 下 prompt，（I'm prompting one agent, or I have an agent that's prompting agents that's prompting agents,）
- `00:48` 就像是一個由數千個 Agent 組成的樹狀結構。（and it's like a tree of thousands of agents.）
- `00:50` 但我認為在處理這些事情時，最重要的觀念就是，每當 Claude 出錯時，（But I think it's just like the most important idea when working on this stuff is like every single time Claude Code makes a mistake,）
- `00:57` 我不會直接告訴 Claude 要怎麼做才對。（I don't tell Claude Code to do it differently.）
- `00:58` 我會叫它把它寫進 CLAUDE.md，或者建立一個 skill 之類的東西來修正做法。（I tell it to write it to the QuadMD, or to like make a skill or something to do it differently.）
- `01:04` 如果你能做到這一點，Claude 就能夠一直運作下去。（And if you can do this, then Claude Code can just like run forever.）
- `01:07` 我認為我們意識到的另一件事是，驗證真的非常重要。（And I think the other thing that we kind of realized is the verification is really important.）
- `01:11` 我們之前並沒有意識到這一點。（Like we didn't realize that.）
- `01:12` 我常從我們接觸的開發者和企業那裡聽到這個問題。（I hear this come up a lot with developers and enterprises that we meet with.）
- `01:17` 你有什麼建議能讓 Claude Code 在驗證方面表現得更好嗎？（What are your tips for making a really good, making Claude Code really good at verification?）
- `01:21` 我總覺得這是一件大家都誤解的事情，（I sort of feel like this is this thing that just like everyone misunderstands,）
- `01:24` 因為每當我們談論驗證時，人們想到的都是單元測試，（because whenever we talk about verification, people are thinking like unit tests,）
- `01:28` 或是想到 lint 或型別檢查。（or they're thinking like lint or like type check.）
- `01:31` 這些顯然是非常容易自動化的東西。（These are the things that are obviously really easy to automate.）
- `01:34` 而且這些東西早就已經自動化了。（And these are the things that were already automated.）
- `01:36` 但實際上，當我們談論 Agent 的驗證時，情況稍微有些不同。（But actually, when we talk about verification for agents, it's something slightly different.）
- `01:40` 這更像是：這個 Agent 能執行那個東西嗎？（It's like, can the agent run the thing?）
- `01:42` 這需要花點心思去思考到底該怎麼做，（It takes a little bit of mental work to figure out how exactly do you do this,）
- `01:45` 因為這通常不是那麼直接。（because it's often not straightforward.）
- `01:47` 我認為這就是挑戰之一。（And I think that's like, that's one of the challenges.）
- `01:49` 我記得在 Opus 4 時，Claude 會自我測試。（I remember with Opus 4, Claude tested itself.）
- `01:54` 我們只是把它連接到 Opus 4。（And we just like hooked it up to Opus 4.）
- `01:58` 我當時說：「Claude，建立一個功能，然後在 Bash 裡測試你自己。」（And I was like, Claude, build a feature and then test yourself in like Bash.）
- `02:02` 它打開了一個小型的 Claude CLI 並測試了它自己的功能。（And it opened a little Claude CLI and tested its own feature.）
- `02:07` 我當時心想：「哇，這太瘋狂了。」（And I was just like, whoa, it's crazy.）
- `02:10` 現在我們已經習以為常了。（Like now we're so used to it.）
- `02:12` 現在，你知道，我們有這些循環在執行，像是 iOS 模擬器、（Like now, you know, now we have these loops going for, you know, like the iOS simulator）
- `02:15` Android 模擬器，還有桌面電腦的環境。（and the Android simulator and like computers for desktop.）
- `02:18` 這已經沒什麼好驚訝的了。（Like it's not surprising.）
- `02:20` 但在當時，那真的很瘋狂。（But back then, that was crazy.）
- `02:22` 你們現在是怎麼做的？（How are like, how are you doing it?）
- `02:24` 我最近主要是在開發桌面應用程式。（So I've been mainly hacking on the desktop app these days.）
- `02:27` 團隊中的一名工程師實際上增加了一個桌面開發 skill，（And one of the engineers on the team actually added this desktop development skill）
- `02:31` 教導 Claude 如何執行本地的桌面應用程式。（that teaches Claude how to run the local desktop app.）
- `02:34` 我一直讓它使用這個功能。（And I've been having it use it.）
- `02:36` 它有時還是會遇到問題，或是 staging 環境的錯誤。（And it still runs into issues or like bugs with the staging environment sometimes.）
- `02:42` 所以在這種情況下，我會讓它去讀取 Slack 並理解，（And so what I have it do is in those cases, I have it read Slack and understand,）
- `02:46` 嘿，現在 staging 環境是不是掛了？（hey, is staging down right now?）
- `02:48` 或者是不是有其他人已經遇到過這個問題了？（Or is there, has someone else already hit this?）
- `02:51` 然後當它除錯完整個問題後，我會叫它更新那個桌面開發 skill。（And then when it debugs the whole issue, I tell it to update the desktop development skill.）
- `02:56` 這個 skill 的作用是讓 Claude 啟動一個本地的桌面應用程式。（What this skill does is Claude actually spins up a local desktop app）
- `03:00` 它會使用電腦操作功能在上面點擊。（and it uses computer use to click around on it.）
- `03:02` 所以當我新增一個使用者體驗時，它會點擊並呼叫新的使用者體驗。（And so when I add a new UX, it clicks around to invoke the new UX.）
- `03:07` 它也會測試邊緣案例。（It also tests edge cases.）
- `03:09` 當出現問題時，它會進行修復並重新檢查。（And when there's an issue, it fixes it and rechecks.）
- `03:12` 老實說，這是這個團隊最讓我喜歡的事情之一，就是每個人都會寫程式碼。（This is like, honestly, one of my favorite things about this team is everyone codes.）
- `03:17` 我從未待過一個連產品經理（PM）都會寫程式碼的團隊。（I have never been on a team where my PM would code.）
- `03:23` 這真的很瘋狂。（And it's like crazy.）
- `03:24` 而且你的程式碼寫得非常好。（And your code is really good.）
- `03:26` 你太客氣了。（You're too nice.）
- `03:28` 但我只是覺得這也變得越來越容易了，因為基本上，（But I also just feel like it's also just becoming easier because it's like,）
- `03:32` Claude Code 會負責寫程式碼。（essentially, Claude writes the code.）
- `03:34` 所以更重要的是，你有什麼樣的想法？（And so what matters a little more is like, what's the idea that you have?）
- `03:38` 我覺得如果你是一個具備產品背景和商業背景的人，（And I feel like if you're a person that has like the product context and the business context,）
- `03:42` 並且在思考設計和使用者需求，你自然會想出更好的點子。（and you're thinking about the design and the user, you're just going to come up with better ideas.）
- `03:46` 這就像所有的角色都在融合一樣。（It's kind of like all the roles are merging.）
- `03:48` 我記得剛看到我們設計師 Megan 的 PR 時，我一開始簡直嚇壞了。（I remember seeing Megan, our designer's PRs, and I was just horrified at the beginning.）
- `03:52` 我心想，天哪，為什麼 Megan 會發 PR？（I was like, oh, my God, why is Megan putting up PRs?）
- `03:54` 然後她說，對啊，我只是在修復按鈕。（And then she was like, yeah, yeah, I'm just like, I'm fixing the button.）
- `03:57` 我當時想，好吧，好吧，程式碼看起來不錯。（And I was like, okay, all right, well, the code looks good.）
- `04:00` 所以也許這樣也沒問題。（So maybe it's fine.）
- `04:02` 我覺得現在這已經完全正常了。（And I feel like now it's just like, it's totally normal.）
- `04:04` 沒錯。（Yeah.）
- `04:04` 我們在所有對談的企業中都看到了這種情況。（We see this across all the enterprises we talk with.）
- `04:07` 工程師會先採用 Claude Code。（Like, it's the engineers adopt Claude Code first.）
- `04:10` 然後周邊的角色會在一旁觀察，心想，哇，這東西太強大了。（And then the edge adjacent roles look over their shoulder and they're like, whoa, this thing is very powerful.）
- `04:16` 讓我試試看。（Let me try it out.）
- `04:17` 我們發現，這真的很瘋狂。（And we found, it's crazy.）
- `04:19` 我們發現設計師的生產力更高了，他們可以直接在應用程式中製作原型並進行修改，而不必付錢請工程師幫忙。（We found that like our designers are more productive, making prototypes and making changes directly in the app instead of paying an engineer.）
- `04:26` 產品經理（PM）也在應用程式中進行修改。（PMs are making changes in the app.）
- `04:29` 我們的財務團隊也在用 Claude Code 運作。（Like our finance team runs in Claude Code.）
- `04:32` 他們在那裡進行預測。（They do their projections there.）
- `04:34` 資料科學也是。（Data science.）
- `04:35` 如果你跟我們的資料科學家聊聊，真的很酷。（Like if you talk with our data scientists, it's so cool.）
- `04:38` 每個人螢幕上都開著 Claude Code。（It's just like everyone just has Claude codes up on their screens.）
- `04:42` 我覺得它對不同角色來說都非常萬能。（I feel like it's remarkably versatile for different roles.）
- `04:46` 你覺得現在有哪些使用案例正在挑戰極限？（What do you feel like nowadays are like the use cases that are pushing the limit?）
- `04:51` 我最興奮的一個是常規任務（routines）。（One that I'm super excited about is routines.）
- `04:54` 我們團隊有一位工程師在我們所有的產品中推出了語音模式。（There's one engineer on our team who launched voice mode across all of our products.）
- `04:59` 他設定了一個常規任務，專門監聽每一個進來的工單、每一個 GitHub issue、每一個關於語音模式的錯誤報告。（And he has this routine set up that just listens for every ticket that comes, every GitHub issue, every bug report about voice mode.）
- `05:08` 然後他的 Claude 會自動接收，主動提出修復方案，並將 PR 傳送給他。（And his cloud just picks it up, proactively puts up a fix, and then pings the PR to him.）
- `05:14` 當他在語音模式上成功實現這一點時，他想，好吧，我們收到了很多其他沒有被回應的回饋。（And when he got that working for voice mode, he thought, okay, we're getting a lot of other feedback that isn't being responded to.）
- `05:21` 所以他也設定了一個常規任務來監聽這些回饋。（So he also set up a routine to listen for that.）
- `05:24` 我發布了一個小功能，裡面有一個我沒注意到的邊緣案例。（So I shipped this small feature, and there was like an edge case in it that I didn't see.）
- `05:29` 有人為此提交了一個錯誤報告，我本來打算那天晚上處理這個錯誤。（And so someone filed a bug for it, and I was going to get to the bug that night.）
- `05:34` 當我的 Claude 正在工作時，它說，等一下，另一個 Claude 已經修復了這個問題。（And as my Claude Code was working, it said, wait a second, another Claude Code has already fixed this.）
- `05:39` 我心想，這怎麼可能？（And I was like, how is this possible?）
- `05:40` 我之前從未跟他討論過這個功能。（Like, I've never talked to him about this feature before.）
- `05:42` 所以我傳訊息問他，你是怎麼這麼快修復它的？（And so I pinged him, and I was like, how did you fix this so quickly?）
- `05:45` 他說他有另一個常規任務，專門尋找五小時內未被回應的錯誤報告，並提出修復方案。（And he says he has another routine that just looks for bug reports that haven't been responded to in five hours and puts up a fix.）
- `05:52` 而且他會直接合併那些容易驗證的修復。（And he merges the ones that are easy to verify.）
- `05:55` Claude Code 現在總是這樣告訴我。（Claude Code tells me this, like, all the time now.）
- `05:57` 說已經有其他人修復它了？（That someone else has already fixed it?）
- `05:59` 總是有另一個人的 Claude Code 在處理這件事。（There's always, like, another person's Claude Code that's working on it.）
- `06:01` 沒錯，這就是其中一個改變。（It's like, yeah, that's been one of the changes.）
- `06:04` 我覺得我們在不久前還在試著弄清楚該如何使用 routines。（I feel like we're, a while ago, we were trying to figure out, like, how to use routines.）
- `06:08` 我覺得 Agent SDK 是我們第一個想到可以透過程式化方式使用 Claude Code 的點子。（And I feel like just, like, the agent SDK was this first idea that we could use Claude Code programmatically.）
- `06:15` 但我覺得剛開始時，我們並不清楚該如何使用它，或是該用它來做什麼。（But I feel like at the beginning, it just wasn't obvious how do we use it, what do we use it for.）
- `06:20` 而我認為 routines 是第一個非常顯而易見的應用。（And I think routines are the first really obvious application.）
- `06:24` 我不知道該怎麼說，它就是會處理所有的程式碼審查。（And I don't know, like, it just does, like, all the code review.）
- `06:27` 它會像保姆一樣盯著每一個 PR。（It babysits, like, every PR.）
- `06:30` 你還記得以前嗎？你以前真的必須親自回覆程式碼審查的意見。（You remember back in the day, you used to actually have to, like, respond to code review comments.）
- `06:34` 你以前必須自己修復 CI。（You used to have to, like, fix CI.）
- `06:36` 你以前必須自己進行 rebase。（You used to have to rebase.）
- `06:38` 沒錯。（Yeah.）
- `06:39` 我已經很久沒做這些事了。（Like, I haven't done that in a long time.）
- `06:40` 沒錯。（Yeah.）
- `06:41` 當你在 CLI 中與 Claude Code 同步工作時，你最常用的功能是什麼？（When you're in the CLI and you're synchronously working with Claude Code, what are your go-to features?）
- `06:47` 好吧，以前是 plan mode。（Okay, what they used to be is plan mode.）
- `06:49` 我現在已經不用那個了。（I don't use that anymore.）
- `06:50` 那你現在用什麼？（What do you use instead?）
- `06:52` Auto mode。（Auto mode.）
- `06:53` Auto mode。（Auto mode.）
- `06:53` 這是最棒的。（It's the best.）
- `06:54` 取代了 plan mode？（Instead of plan mode?）
- `06:55` 取代了 plan mode。（Instead of plan mode.）
- `06:56` 對，因為較新的模型其實不再需要規劃步驟了。（Yeah, because the newer models, they don't actually need, like, a planning step anymore.）
- `07:00` 我認為這對於 Opus 4 到 4.5 左右的版本來說真的很重要。（I think this was really important for, like, Opus 4 through maybe 4.5.）
- `07:05` 但我覺得從 4.6 開始，絕對到了 4.7，它就不再需要那個規劃步驟了。（Then I think starting with 4.6 and definitely with 4.7, it just doesn't need that planning step.）
- `07:09` 我想有些人還是會用它。（I think some people still use it.）
- `07:10` 他們喜歡保留那個 artifact。（They like to have that artifact.）
- `07:11` 但我不用。（I don't use it.）
- `07:12` 我做任何事都直接用 auto mode。（And I just do auto mode for everything.）
- `07:14` 因為這樣我啟動 Claude Code 後。（Because then I start my Claude Code.）
- `07:16` 它就開始工作了。（It starts to work.）
- `07:17` 然後我就可以直接處理下一個 Claude Code 任務。（And then I just, like, move on to the next Claude Code.）
- `07:19` 我不需要坐在那裡盯著它看。（And I don't have to sit there and watch it.）
- `07:20` 但在早期階段，我們對 Claude Code 有這種權限提示模型，對吧？（But from the very early stages, we had this, like, permission prompts model for Claude Code, right?）
- `07:25` 它會執行一個工具，然後問你：嘿，你確定要執行這個工具嗎？（Like, it runs a tool and then it asks you, like, hey, are you okay running this tool?）
- `07:29` 而你必須回答是或否。（And you had to say yes or no.）
- `07:31` 在一年半前，那已經是我們能做到的極限了。（And at the time, that was kind of the best we had a year and a half ago.）
- `07:34` 因為當時我們還沒有分類器。（Because we didn't have, you know, classifiers.）
- `07:36` 模型的對齊程度也沒有今天這麼好。（The model was not as well aligned as it is today.）
- `07:38` 所以 auto mode 真的是一大進步。（So auto mode was just such a big step up.）
- `07:41` 因為實際上，你並不想閱讀大部分的請求。（Because actually, you don't want to read most of these requests.）
- `07:44` 將請求路由到不同的模型並讓它進行安全性檢查，效果好多了。（Just routing it to a different model and having it check for security works so much better.）
- `07:49` 如果某件事看起來有點可疑，或者不是你認為你想執行的指令，又或者是不安全的，模型就會直接拒絕它。（And if a thing, like, is a little sus or, you know, this isn't a command that you think you want to run or it's not safe, the model will just deny it.）
- `07:57` 然後你可以回頭再允許它。（And then you can go back and you can allow it later.）
- `08:00` 我認為這是那種階段性的改變之一。（I think this has been one of those, like, step changes.）
- `08:02` 我們在一年半前根本不可能做到這一點。（We just, there's no way we could have done this a year and a half ago.）
- `08:04` 當你接受了 99% 的請求時，人類的天性就是閱讀時會自動忽略內容。（It's just human nature when you accept 99% of requests that your eyes just glaze over when you read it.）
- `08:11` 所以實際上，我們認為自動模式比閱讀每一個權限提示更安全。（And so actually, we feel that auto mode is more safe than reading every single permission prompt.）
- `08:16` 因為這意味著你只需要專注於最重要的事情。（Because it means that you're only paying attention to the most important thing.）
- `08:20` 而不是被那些 99% 都是同意的訊息給淹沒。（And not, like, being spammed a bunch of things that are just 99% yes.）
- `08:24` 我認為安全性就是這樣，你可以談論它，但要正確地執行它又是另一回事。（I think security is one of these things, like, you can talk about it and then it's a totally different thing to actually do it correctly.）
- `08:30` 因為它看起來並不總是像你想像的那樣。（Because it just doesn't always look the way that you think it's going to look.）
- `08:33` 重點在於始終進行紅隊測試、始終進行滲透測試、始終保持觀察，你知道的，始終擁有威脅模型，然後利用它來找出這個東西會如何被攻擊？（And it's just all about, like, always red teaming, always pen testing, always looking, you know, always having a threat model and then using that to figure out, you know, how is this thing going to get attacked?）
- `08:43` 人們會如何進行 prompt injection？（How are people going to get prompt injected?）
- `08:44` 沒錯。（Exactly.）
- `08:45` 我感覺團隊對此非常執著。（And I just feel like the team is just, like, obsessed with this.）
- `08:48` 這非常重要，因為結果就是，我信任這個 Agent 去執行任務。（And it's so important because, as a result, I just trust the agent to run.）
- `08:53` 我可以繼續處理其他事情，甚至可以使用第二個 Agent。（And I can move on and I can just have, like, a second agent.）
- `08:57` 如果我不信任它，我就無法做到這一點。（And if I didn't trust it, then I just wouldn't have been able to do that.）
- `09:00` 在內部，為了將自動模式推向我們的使用者，我們必須先真正信任它。（And internally, to actually get auto mode out to our users, we needed to really trust it first.）
- `09:07` 所以我們收集了數千份完整的 Agent 軌跡和權限提示的轉錄內容，並讓自動模式來分類它是否安全。（And so what we did was we collected thousands of transcripts of, like, an entire agent trajectory and a permission prompt and had auto mode classify whether or not it was safe.）
- `09:18` 而它在這方面的表現極其出色。（And it was extremely good at this.）
- `09:20` 接著我們找來紅隊測試人員，請他們嘗試進行 prompt injection 並試圖駭入程式庫。（So then we got red teamers and we asked them to try to prompt inject and try to hack the code base.）
- `09:26` 我們利用這些結果來建立評估基準，並確保所有這些攻擊都被拒絕。（And we used this to create evals and made sure that all of these were denied.）
- `09:30` 然後我們讓內部的團隊嘗試進行 prompt injection 並駭入 Claude Code 的自動模式。（And then we had our own internal teams try to prompt inject and hack Claude Code, Claude Code's auto mode.）
- `09:37` 接著我們改進了自動模式，確保我們能攔截所有這些攻擊。（And then we improved auto mode to make sure that we caught all of these.）
- `09:40` 所以這不僅僅是保護你免受當今野外環境中存在的漏洞影響，更是防禦我們所能建構出的最聰明的攻擊。（So it's not only just protecting you against the vulnerabilities that are out there in the wild today, but the most intelligent attacks that we can construct.）
- `09:49` 是啊，老實說這是一種很奇特的方法。（Yeah, I mean, it's like, it's honestly like a weird approach.）
- `09:52` 我覺得過去一年裡有各種功能，當有人第一次提出時，我心想：「啊，不，這絕對行不通。」（I feel like there's like all these features the last year where the first time someone pitched it, I was like, ah, no, no way that's not going to work.）
- `09:58` 這絕對行不通，但我覺得隨著時間推移，我發現我其實錯了。（way that's not gonna work and i feel like over time i've just learned like i'm actually wrong）
- `09:59` 而且我感覺這種情況發生得太頻繁了。（And I feel like over.）
- `10:01` 因為在模型上進行建構真的很奇特，沒錯，這就像是所有這些...（like so often now because like building on the model is so weird yeah it's just like all this）
- `10:07` 我多年來學到的工程知識，其中很大一部分我必須拋棄。（like engineering stuff that i've learned over the years like so much of it i just have to like）
- `10:11` 這就是現在工作的一部分，我們正在基於新事物進行建構，（throw out and this is just like part of what the job is now like we're building on a new thing and）
- `10:14` 我們必須重新學習。自動模式絕對是其中之一，第一次聽到時，我心想：（we just have to relearn it and i auto mode was definitely one of these i was like the first time）
- `10:19` 「將 prompt 路由到模型？沒門，這行不通。」但結果證明，從實證來看，它運作得非常好。我聽說你也喜歡 Loop。（i heard it i was like route the prompt to a model no way that's not gonna work and then it actually）
- `10:23` 嗯，沒錯，我喜歡 Loop。（turns out empirically it works really really well i heard you also love loop um i yeah i love loop）
- `10:29` 你是怎麼使用它的？我認為對於 Loop，我們在一年半前經歷了一次轉變，當時我們想：「好吧，有原始程式碼，但工程師真正應該互動的對象可能不是原始程式碼，而是 Agent。」（how do you use it i think for loop there's this transition that we went through like a year and）
- `10:35` 所以我們跨出了這一步：（a half ago where we were like all right there's source code but actually the thing an engineer）
- `10:41` 我不寫原始程式碼，我與 Agent 對話，由 Agent 為我編寫原始程式碼。（should interact with maybe it's not the source code maybe it's the agent and so we made this leap of）
- `10:46` 我認為現在正在發生的事情是我們正在邁出下一步：（like i don't write the source code i talk to an agent and the agent writes the source code for me）
- `10:51` 我不再與 Agent 對話，我與 Loop 或 Routine 對話，它會為我向 Claude Code 發送 prompt，這真是太瘋狂了。（and i think right now what's happening is we're making the next leap i don't talk to an agent）
- `10:56` 我的意思是，這才過了一年半，如果退一步看，這已經是兩次巨大的飛躍了。（anymore i talk to a loop or i talk to a routine and it prompts Claude Code for me and it's just it's crazy）
- `11:04` 你是如何看待整個工程組織的變革的？我要戴上我的商業帽了。（i mean it's been like it's a year and a half and this is like two big leagues if you take like a step）
- `11:08` 我有一個最喜歡的案例研究，這是 90 年代的《哈佛商業評論》。（back how are you seeing entire engineering orgs change i'm gonna put i'm gonna put on my business）
- `11:13` 他們當時在討論：「電腦已經出現了，為什麼我們還沒看到生產力提升？」（cat hat i i have this like favorite case study this is like a harvard business review from the 90s）
- `11:18` 這是一個非常精彩的快照，記錄了當時人們的真實感受。（and they were talking about like computers are here why are we not seeing the productivity benefits）
- `11:22` 因為你知道，人們過去使用大型主機，後來公司轉向個人電腦，這算是一種新事物，公司正在試圖弄清楚如何使用它，（and it's just this like amazing snapshot into like what it actually felt like at the time）
- `11:27` 就像他們現在試圖弄清楚如何使用 AI 一樣。結果發現，要從電腦中獲得生產力提升，你不能只是保留你的紙本檔案櫃和紙筆業務流程，然後在旁邊放一台電腦做點什麼。（because like you know people used to use mainframes at some point companies switched to personal）
- `11:31` 實際上，你必須做的是丟掉檔案櫃，丟掉所有的紙和筆，然後把電腦放在中心，一切都必須透過電腦運作。（computers it was sort of a new thing and companies were trying to figure out how to how they use it）
- `11:35` 它必須成為每個業務流程的核心。我覺得在 Anthropic，我們在做這件事，當你入職時，你不會問人們問題，因為沒人在他們問我問題時...（the same way they're trying to figure out how to use ai right now and it turned out that to get the）
- `11:40` 電腦帶來的生產力效益，並不是說你原本有紙本檔案櫃，（productivity benefits from computers what you had to do isn't like you have your paper filing cabinet and）
- `11:46` 還有紙筆作業流程，然後旁邊放一台電腦來做點什麼，（your like paper and pen business process and then there's like a computer on the side that does）
- `11:50` 實際上你必須做的是丟掉檔案櫃，丟掉所有的（something actually what you have to do is you throw out the filing cabinet you have to throw out all your）
- `11:55` 紙張和筆，然後把電腦放在中心，所有事情都必須透過（paper and all your pens and then you put a computer in the center and everything has to run through the）
- `11:59` 電腦來執行，它必須成為每個商業流程的核心。我覺得在 Anthropic 我們（computer it has to be at the center of every business process and i i feel like at anthropic we do）
- `12:03` 做了一件事，就是當你入職時，你不會問別人問題，沒人會在我（this thing where when you onboard you don't ask people questions like no one asks me questions when they）
- `12:09` 入職時問我問題，你可能也一樣，他們會問 Claude，這有點奇怪，這是我（onboard you probably have the same thing they ask Claude Code and this is kind of weird like this is the first）
- `12:15` 待過的第一家這樣的公司，我覺得對我們來說，Claude 就是一切的核心，（company i've been at like that and i feel like for us Claude Code is just at the center of everything）
- `12:19` 每當我有問題，我就問 Claude；每當我寫程式碼，我就用 Claude；每當我需要程式碼審查，（whenever i have a question i ask Claude Code whenever i write code i use Claude Code whenever i need a code review）
- `12:23` Claude 就會做；每當我需要安全性審查，Claude 就會做；每當我需要，你知道，填寫表格（Claude Code does it whenever i need a security review Claude Code does it whenever i need to you know fill out a form）
- `12:29` 之類的，Cowork 就會做，所以 Claude 就是一切的核心。我覺得（or something co-work does it so it's just like Claude Code is at the center of everything and i i feel like）
- `12:34` 那些真正搞懂這點的公司，現在已經有一堆了，他們只是把（the companies that are really figuring it out and there's a bunch of them now they're just putting）
- `12:38` Claude 放在核心位置。但我想對於電腦來說，轉型花了 10 到 15 年，但實際上（Claude Code at the center of it but i think for computers the transition took 10 to 15 years but actually）
- `12:43` 對於 AI 來說，因為我們的工作大部分已經數位化，且 Claude 可以使用電腦，（for ai because so much of our work is already ready digitized and Claude Code can use the computer）
- `12:49` 它能寫程式碼並執行程式碼，這種轉型發生得快得多。我覺得這真的（and it can write code and run code this transition is happening a lot faster i think it's just like）
- `12:54` 非常令人興奮，我覺得現在我不需要再麻煩別人了，當我（really it's just really exciting like i feel like now i don't have to bug people anymore and when i）
- `13:00` 與人互動時，是因為這很有趣，我可以與他們合作處理事情，我們可以（interact with people it's because it's like fun and i get to collaborate with them on stuff and we get）
- `13:05` 一起創造些什麼，而不是因為我需要他們，我需要從他們那裡得到什麼，（to create something together it's not that like i need them i need something you know from them）
- `13:10` 因為 Claude 現在真的可以做很多那類事情。我也覺得身為工程師，我（because like Claude Code can actually do a lot of that stuff now and i also feel like as an engineer i've）
- `13:14` 從未像現在這樣享受工程開發的樂趣，因為那些繁瑣的部分我（just never had this much fun doing engineering because they're like the tedious part i don't）
- `13:18` 不需要做了，我只需要提出想法，與客戶交談，每一個我不喜歡的想法，我（have to do like i'm just coming up with ideas i'm talking to customers and every idea i don't like i）
- `13:24` 不再需要待辦清單了，因為 Claude 直接幫我完成所有事，所以我的工作就是提出（don't have a to-do list anymore like Claude Code just builds everything and so my job is to come up with）
- `13:28` 這些想法，這真的太有趣了。好，這裡有個問題：未來是產品還是工程？（these ideas and it's just so fun okay so here's a question is the future product or engineering like）
- `13:33` 每個人都會成為產品經理，還是每個人都會成為工程師？每個人都會兩者兼具。（is everyone going to be a pm or is everyone going to be an engineer everyone's going to be both）
- `13:37` 我強烈感覺到這些角色正在融合。當我們看我們的團隊時，我們的產品（i i feel pretty strongly that these roles are merging like when we look at our team our product）
- `13:43` 團隊都會寫程式碼，我們的開發者關係團隊都會寫程式碼，我們的設計團隊都會寫程式碼，然後我們（team all rights code our dev role team all rights code our design team all rights code and then we）
- `13:50` 看我們的工程師，他們很多人都以「端到端」的方式交付產品，他們有想法知道（look at our engineers and a lot of them ship products end to end they have an idea for what）
- `13:55` 要建構什麼，他們就去建構，他們與法務和行銷合作，弄清楚我們如何將其傳達給（to build they build it they work with legal and marketing to figure out how we communicate this to）
- `14:01` 世界，並確保它是安全的，也包含安全性。很多時候他們只是親自監督（the world and make sure it's safe and with security too and a lot of times they just see through this）
- `14:06` 整個「端到端」的流程。所以我認為現在 AI 真的造福了那些充滿好奇心、（whole process end to end so i think right now ai really benefits people who have a lot of curiosity）
- `14:13` 有產品品味、熱愛這種「端到端」所有權的人。現在很多人（how a lot of product taste who love to have this like end-to-end ownership and now a lot of people）
- `14:19` 正在運行數百個 Agent，你認為人們在從單一 Agent 過渡到多個，甚至數百個 Agent 時，應該採用什麼產品？（are running like hundreds of agents what are the products that you think people should be adopting as）
- `14:25` 直到最近，我寫程式碼的方式是打開（they transition from single to multiple to hundreds until recently the way that i wrote code was i had like）
- `14:32` 六個終端機分頁，裡面有六個相同儲存庫的 git checkout，然後我就在它們之間切換。（six terminal tabs with six git checkouts of the same repo and then i would just like tab between them）
- `14:38` 現在完全不同了，我只有一個分頁，我使用我們剛發布的全新 Agent 視圖，它真的（now it's pretty different i have like one tab i use the new agent view that we just shipped it's like）
- `14:43` 太棒了，我很高興我們花了一些時間來迭代，讓它變得如此好。（so good and i'm so glad that we took a while to iterate on it to make that really good）
- `14:48` 而且我也使用桌面應用程式，因為這樣我就不必處理 checkout 的瑣事，它只是（and i also use the desktop app because i don't have to fiddle with checkouts that way it just like）
- `14:53` 你知道，它會進行工作樹複製，它會為我建立工作樹。（you know it does the work tree cloning like it creates the work trees for me）
- `14:58` 而六個月前我沒想到的一點是，現在我大概有一半的工程工作（and the thing that i would not have expected six months ago is probably half my engineering now）
- `15:03` 不需要我的電腦，因為我運行了太多 Agent，我可以直接從手機開始，（i don't have my phone so i just have like i have so many agents running that i just start from my）
- `15:08` 我使用遠端控制，這現在簡直太神奇了，我會在電腦上啟動某個東西，（phone i use a remote control which is like amazing now and like i'll start something on my computer and）
- `15:14` 然後我就用手機遠端控制它，我可以四處走動，去買杯咖啡，（then i'll just remote control in for my phone and i'll just like walk around i'll like get coffee and）
- `15:18` 然後檢查我的 Agent，也許再啟動另一個 Agent。有時我在跟（then i'll check in on my agents and maybe i'll start another agent and sometimes i'm like talking to）
- `15:23` 某人聊天，我們想到一個新點子，我就會當場啟動一個 Agent，我會跟它（someone and we come up with a new idea i'll just start an agent on the spot i'll like talk to it）
- `15:27` 用語音模式對話，讓它直接建構東西，我甚至不需要回到我的電腦前了。（with voice mode and just have it build something and i don't even have to go back to my computer）
- `15:31` 我記得你開始這樣做的時候，因為你真的會離開公司，把你的（anymore i remember when you started doing this because you would actually leave work have your）
- `15:36` 電腦留在桌上，開著機、插著電、螢幕鎖定，我原本以為你會在某個時間點（computer on your desk open plugged in screen locked and i just thought you would like come back）
- `15:41` 回到辦公室拿你的電腦，但後來時間已經很晚了，我想說（to the office at some point to get your computer but then it would be like pretty late and i was like）
- `15:45` 嗯，也許你只是不小心把它留在這，結果隔天又發生了，（hmm maybe you just like left it here by accident and then it happened again the next day）
- `15:50` 又發生了一次，我就想說，等等，這太奇怪了，因為你還在提交 PR，所以...（and it happened again the next day and i was like wait this is so weird because you're landing pr so）
- `15:55` 你的電腦就在我旁邊，我記得你當時回應說，沒錯，我正從沙發上進行程式開發（your computer is right next to me and i remember you responding and you're like yeah i'm coding from）
- `16:00` 沒錯，那週遠端控制功能變得非常強大，對了，另一個使用者一直在詢問的問題是（my couch yeah that was the week that remote control got really good yeah so another thing that users are）
- `16:06` 如何在大型企業中進行上下文工程，特別是這件事，你知道的，人們過去常談論 prompt engineering，也常談論上下文（asking about all the time is how do you do context engineering especially in a large enterprise this is）
- `16:12` 工程，這在某種程度上與模型當時的能力相符，回到 Sonnet（the thing you know people used to talk about prompt engineering they used to talk about context）
- `16:15` 3.5 的時代，你必須進行 prompt engineering，回到 Opus 4 的時代，你必須進行上下文工程，但隨著（engineering this is sort of matching where the model was at the time back in the days of sonnet）
- `16:21` 在 Opus 4 的那個年代，你必須進行 prompt 工程，你必須進行 context 工程，但現在有了（3.5 you had to prompt engineer back in the days of opus 4 you had to context engineer but with the）
- `16:27` 現在的模型，你不需要做這些，你只需要給它最精簡的系統 prompt，最精簡的（models of today you don't do any of this you give it the minimal possible system prompt the minimal）
- `16:32` 工具，然後讓模型自己去摸索。就像你只需要給模型某種方式（possible tools and then you let the model figure it out like you just have to give the model some way）
- `16:37` 來引入背景資訊，我認為這是最重要的。你是怎麼想的？我的看法（to pull in the context i think that's the most important thing how do you think about it i see things）
- `16:41` 非常相似，我是背景資訊極簡主義者，所以我的基本哲學是只告訴模型它（very similarly i'm a context minimalist so my general philosophy is tell the model only what it）
- `16:48` 需要知道的資訊，剩下的讓它自己去摸索。我認為當你給模型太多背景資訊時，（needs to know and let it figure out the rest of it i think when you give the model too much context）
- `16:55` 這有點像是在微觀管理它，有時模型知道更好的方法來達成（it's kind of like you're micromanaging it and sometimes the model knows a better way to get to）
- `17:01` 相同的結果，我個人傾向於給模型這種自由去發揮。而且總體來說，（the same outcome and i personally prefer to give the model that freedom to do that and then in general）
- `17:07` 我們也正在讓我們的 harness 更精簡，這樣你就有更多空間來輸入自己的 prompt，也能讓它（we're also making our harness more lean so that you have more room for your own prompts and so that）
- `17:13` 更好地遵循你的 prompt。現在有各種不同的 Agent 方式，但我覺得在（follows your prompts better there's all these different ways to squad now but i feel like in a）
- `17:18` 一年後，這將會是一套全新的東西，如果還是一樣的東西會讓人非常驚訝，（year it's going to be a totally new set of things and it's going to be so surprising if it's still these）
- `17:23` 因為我覺得我們現在正看到這些巨大的趨勢正在發生，Agent 正在（same things because i i think like we're seeing these giant trends happening right now agents are）
- `17:28` 運行更長的時間，它們更具自主性。我很少一次只運行一個 Agent，通常是（running for longer they're more autonomous very rarely am i running one agent at a time it's usually）
- `17:34` 幾個 Agent，或是幾十個、幾百個甚至幾千個，所以它的外觀形式（like a few agents or dozens or hundreds or thousands and so like the form factor for that）
- `17:38` 將會與過去的截然不同，我不知道會是什麼樣子，而且我（it's going to be really different than what came before and i don't know what it's going to be and i）
- `17:43` 認為很大程度上這將取決於團隊去摸索出來。這就是為什麼我（think in a large part it's going to be up to the team to figure it out and this is um this is why i'm）
- `17:48` 很高興我們以這種方式經營團隊，每個人都能提出想法，（like so happy we run the team that the way that we do where everyone just comes up with ideas and）
- `17:52` 每個人都能思考產品，每個人都隨時與使用者交流，因為我不認為（everyone is able to think about the product everyone talks to users all the time because i don't think）
- `17:57` 這些想法會來自我們，它們會來自團隊，完全來自與我們一起開發的（these ideas are going to come from us it's going to come from the team totally and from everyone in our）
- `18:02` 社群中的每一個人。（community building with us）

## 標籤

Claude Code, Agent, 功能更新, 自動化, CLI, Anthropic, Claude