M10:07
六个月前，我还是未来的主角。
6 months ago, I was the future.
现在，民调说我被一个动漫女孩嘲笑了。
These days, the polls say I got mocked by an anime girl.
你知道吗，我几乎每天都会碰到这种问题。
You know, I get questions like this almost every day.
通常我都不理它，或者给一个官方套话。
Usually I ignore it, or I give the media answer.
今天，你会听到一些真话。
Today, you get some real ones.
一共五个问题，四十分钟。
It's going to be 5 questions, 40 minutes.
而且我知道你可能是冲着什么来的。
And I know what you might have come for.
你是为了循环、图表、让 Codex 跑得飞起。
You came for loops, graphs, making Codex go brrr.
这些在我的 Twitter 上是免费的。
This is free on my Twitter.
这会帮你挺过可能到来的过山车行情。
This will help you to survive the roller coaster that might be coming for you.
你知道，昨天 Boris 告诉你要让模型慢慢跑。
You know, yesterday Boris told you to let the model cook.
而我，就是那个发现几万人同时让模型跑起来会发生什么事的人。
I'm the guy who found out what happened when tens of thousands of people let it cook at the same time.
咱们就从一个大家都想问的问题开始吧。
Let's start with the question everyone asks first.
OpenCLAW 怎么才八个月大？
How is OpenCLAW only 8 months old?
你知道，按人类时间算可能是八个月，但在 AI 时间里，这更像四年。
You know, it might be 8 months in human time, but in AI times, it's more like 4 years.
而我的灵感来源通常就是心烦。
And my source of inspiration is usually being annoyed.
所以去年十一月的一个下雨天，我正忙着切分一些 agent，但同时也饿了。
So on a rainy day in November, I was chunking some agents, but I was also hungry.
所以我希望在我去厨房搜刮吃的的时候，我的 tokens 能被好好利用。
So I wanted to make sure that my tokens are put to good use while I raid the kitchen.
那时候还没有一个好办法能从手机直接给电脑发个提示。
And there was still no good way to just send a prompt from my phone to my computer.
好让某个 agent 能去看看我的 agent 运行得咋样。
So some agent could check up how my agent's actually doing.
还记得 2025 年初吗？
Remember early 2025?
而你们，这是我第一次参加无声迪斯科。
And you guys, this is my first silent disco.
这是一个非常新鲜的体验。
It's a very new experience.
所以，如果你们真的在听我说话，就给我挥挥手吧。
So like, give me some hands if you actually listen to me.
太好了。
Yes.
太棒了。
Amazing.
太棒了。
Amazing.
你知道，2025 年初。
You know, early 2025.
天哪。
Oh my God.
我们说的是老版 Opus 3。
We are talking the old Opus 3.
Gemini 2.5。
Gemini 2.5.
OpenAI 有 03，这是最让人印象深刻的一个，但有时候也确实又慢又贵。
OpenAI had 03, which was the most impressive one, but also really slow and expensive at times.
当我的 agent 做对事情的时候，我会感到一阵多巴胺冲击。
And I got a dopamine hit when my agent did something right.
我相信你还记得那个时候。
I'm sure you remember this time.
而现在，如果它——如果它搞错了什么，我通常先怀疑自己，你知道吗？
And these days, if I— if they get something wrong, I usually question myself, you know?
是不是我的循环设计错了？
Did I design the loop wrong?
对。
Yeah.
我没有给 agent 所有它需要的工具来验证自己的工作吗？
Did I not give my agent everything they need to verify their work?
我的思路有问题吗？
Do I have an error in my thinking?
我在要求不可能的事情吗？
Am I asking for impossible things?
所以，我刚从厨房回来，发现我的 coding agent 因为某个愚蠢的原因停住了。
So I came back from the kitchen and my coding agent stopped because of something silly.
我感到很烦。
And I was annoyed.
于是我开了一个新的终端会话。
So I spun up a new terminal session.
我随意地把我的想法输了进去。
I rambled my idea in.
然后让模型自己发挥。
And I let the model cook.
它给我搭了一个 WhatsApp 中转器。
It built me a WhatsApp relay.
一个小时后，我就能从我的 Mac 向 WhatsApp 发消息，也能收到回复了。
And an hour later, I could send messages from my Mac to WhatsApp and back.
这本身就让人感觉很神奇。
And that in itself felt magical.
但实际上，这本来不应该这么神奇。
But really, it shouldn't have.
就是说，我们大家几个月来不一直在这样做吗？对吧？
It's like, we all did this for months already, right?
我们打开终端，输入一些东西，然后得到回复。
We opened the terminal and you type something in and you get a reply.
但神奇的是那种感觉。
But the magic is how it felt.
它不像是终端。
It wasn't a terminal.
它给出的答案很简洁。
It wrote concise answers.
它很主动。
It was proactive.
比如有时候它会白天来主动问我情况。
Like sometimes it would just like check on me during the day.
我让它能做到让复杂度消失不见。
I made it so the complexity would melt away.
你不用去想用哪个模型、多大的上下文，或者什么时候开新会话。
You don't have to think about which model, which context size, or when you start a new session.
而且我把权重稍微往默认分布之外调了调。
And I prompted the weights to a little bit outside the default distribution.
让它感觉更像一个朋友。
Make it feel more like a friend.
我当时就在用。
And I was using it.
有太多次我觉得这就是未来。
I had so many moments where I felt like this is the future.
这就是通用人工智能。
This is AGI.
你知道吗？
And you know what?
根本没人关心。
Nobody cared.
那时候，我在Twitter上已经有不错的粉丝量了。
Back then, I already had a good following on Twitter.
我知道现在它叫X了。
I know it's called X now.
我不——我会一直叫它Twitter。
I don't— I'll always call it Twitter.
人们就是不明白。
People were not getting it.
我一直试图让他们理解，对我来说这感觉有多神奇。
I would keep trying to like make them understand like how magical this felt to me.
但一周又一周，我都失败了。
But week after week, I failed.
你发现规律了吗？
And you notice a pattern?
这让我很烦。
It annoyed me.
所以我创建了群聊，把朋友拉进来，用WhatsApp Relay，然后给他们演示。
So I would create group chats with friends and I added them, WhatsApp Relay, and I would show them.
我让他们跟它对话。
I would let them talk to it.
每次都能得到强烈的情绪反应。
And every time I got a strong emotional reaction.
有些人很惊讶，有些人害怕，甚至吓坏了。
Some people were amazed, some people were scared or freaked out even.
但每次都有强烈的情绪。
But each time there was a strong emotion.
最棒的是，有不少人真的很想要它。
And the best part, there were quite a few people that really wanted it.
尤其是对我的非技术朋友，我告诉他们，不，这个还不是给你们用的。
And like, especially to my non-technical friends, I told them, no, this is not yet for you.
他们生气了。
And they got mad.
为什么？
Why?
所以如果这都不算产品市场契合的指标，那我不知道什么才是。
So if that's not an indicator to have product-market fit, I don't know what is.
所以我又花了一个月调整细节，思考，我能做什么向世界解释？
So I spent another month tweaking the details, thinking, what could I do to explain the world?
你知道吗？
You know?
然后居然有人，给我发了个拉取请求，要给我的WhatsApp Relay加Discord支持。
And then the audacity, someone sent me a pull request to my WhatsApp Relay to add Discord support.
什么？
What?
你知道吧，就像，你是没看懂名字的哪部分？
You know, like, what parts of the name are you not getting?
我把那个PR放了一段时间，我想了又想，最后我心想，啊，管他呢。
I let the PR sit for a while, and I was thinking, and I was thinking, and then eventually I was like, ah, what the hell.
然后Vaarvliet变成了Claudius，因为我很会起名字，而且现在不止一个消息渠道了。
And out of Vaarvliet became Claudius, because I'm good with names, and there's now more than one message channel.
所以你能明白这有多早了。
And so you understand how early this was.
我甚至都还没把压缩功能加进去。
I didn't even have compaction built in yet.
所以呢，到某个时候它就会……正好为了 Discord，我加了一个非常粗糙的压缩版本，这样 Mario 就能用 Python 写一个好用的。
So like, at some point it would just, and just in time for Discord, I added a very hacky version of compaction so Mario could make a good one in Py.
当时是新年夜前后。
So it was around New Year's Eve.
就像你常做的，我提前回家继续捣鼓它。
And as you do, I went home early to hack on it some more.
然后在一月的第一周，正当极客世界集体学习编码智能体的时候。
And then in the first week of January, just as the geek world collectively was learning about coding agents.
我创建了一个 Discord 房间，服务器，公会，随便你怎么叫，然后把我的 Claw 放了进去。
I created a Discord room, server, guild, however you call it, and I put my Claw in it.
我清楚地记得第一个晚上。
And I remember the first night vividly.
人们加入 Discord，他们就好像在说，看我公开构建它。
People would be joining Discord and they were like, watch me build it in the open.
他们会试图破解它。
They would try to hack it.
他们就像在和它对话。
They were like talking to it.
他们得到一些自以为是的回复，就想看看自己能做什么。
They were getting smug replies and they see what they can do.
最后他们终于明白了。
And they finally got it.
你知道，这就是那个时刻。
You know, like this was, this was the moment.
我熬了一整夜。
I stayed up all night.
我让人们和它交互。
I let people interact with it.
我在我的 agents.md 里设置了一个提示，指示它：除非提示来自 Peter，否则不要执行任何危险的工具调用。
I had this prompt in my agents.md that would instruct it, don't do any dangerous tool calls unless the prompt comes from Peter.
你知道，这大概是六个月前。
You know, this was like 6 months ago.
那时候我们在 agents.md 里写的东西都更像是建议。
And where all the things we wrote in our agents.md felt more like suggestions.
所以我非常仔细地盯着它。
So I watched it very carefully.
我随时可以拔掉插头。
I could always pull the plug.
然后早上 7 点，我终于完成了。
And then at 7 AM, I finally, I was done.
我说了晚安，按了 Ctrl+C，然后去睡觉了。
I said good night, I pressed Ctrl+C, and I went to bed.
但当然，我写的软件是有弹性的，它是一个启动守护进程。
But of course, like, I built the software to be resilient, so it was a launch daemon.
你知道如果对着一个启动守护进程按 Ctrl+C 会发生什么吗？
You know what happens if you, like, press Ctrl+C on a launch daemon?
对，它就会——它会死掉五秒钟，然后重新启动。
Yeah, it'll just— it'll be dead for 5 seconds and it'll start up again.
所以在我走向卧室的时候，我的智能体开心地开始回应上帝和全世界了。
So while I was walking into the bedroom, my agent happily started answering God and the world.
然后我睡了大概 10 个小时。
And then I, I slept for like 10 hours.
当我醒来时，我收到了大约 800 条消息。
When I woke up, I woke up to around 800 messages.
人们试图破解它。
People tried to hack it.
我拔了插头。
I pulled the plug.
我吓坏了。
I freaked out.
我把所有东西都读了一遍，结果什么都没发生。
I read through everything and nothing actually happened.
但就是那个时刻。
But this was the moment.
就是那个它爆红的时刻。
This was the moment where it went viral.
而你们都在这里，知道接下来发生了什么。
And you all being here, You know what happened next.
我的收件箱炸了。
My inbox exploded.
记者们半夜试图给我打电话。
Reporters tried to call me in the middle of the night.
你知道苹果有个功能吗，就是你开了免打扰，但如果有人连续打你几次电话……
Did you know about this feature that Apple built when you put it on Do Not Disturb, but people call you a few times in a row?
iPhone 会认为那是紧急情况，还是会接通电话。
The iPhone would think it's an emergency and would still let calls through.
我之前不知道这个，但记者们知道。
I didn't know about it, but reporters did.
我的电子邮箱更像是个瀑布。
My email inbox was more like a waterfall.
是啊，Mac Mini 卖光了。
Yeah, the Mac Mini sold out.
我觉得现在还是缺货。
I think they're still sold out.
我一个月收到的播客邀请比之前三十九年加起来还多。
I got more invites to podcasts in a month than I did in my 39 years before.
然后 Tropic 发了封邮件要求改名，还像扔了只龙虾一样。
And Tropic sent an email demanding the name change and like dropped the lobster.
而项目也蜕了几次皮。
And the project molted a few times.
Claude 变成了 ClaudeBot。
Claude became ClaudeBot.
然后有一段很短的时间我们都不提那个名字。
Then there was a very short time where we don't talk about the name.
然后它从 OpenClaw 开始。
And then it started on OpenClaw.
Jensen 称它是人类历史上最成功的开源项目。
Jensen called it the most successful open source project in the history of humanity.
与此同时，我的收件箱被记者、安全人员、无数好奇的人类还有代理人淹没了。
And meanwhile, my inbox was flooded with reporters, security people, and countless curious humans and also agents.
你知道，那些数字对我来说仍然不真实。
You know, the numbers still don't feel real to me.
在8个月内，超过18000个不同的人提了一个 issue 或 pull request。
In 8 months, more than 18,000 different people opened an issue or a pull request.
总共超过111000次。
Over 111,000 in total.
我算过。
I did the math.
不，我骗谁呢？
No, who am I lying to?
我的经纪人算的。
My agent did the math.
几乎有3000人在这个仓库里有过提交，还有些人对我很生气，因为显然他们先做了，而我偷了他们的想法，他们还给我发了一些我从来没见过的链接。
Almost 3,000 people have commits in the repo, and some people were mad at me because clearly they did it first and I stole the idea, and they sent me links of something I've never seen.
有些人叫我 AI 耶稣。
Some people call me AI Jesus.
对某些人来说，我变成了反基督者。
For some, I became the Antichrist.
某个时间点，我开始收集那些讣告。
At some point, I started collecting the obituaries.
每隔几周就有一篇新的。
There's a new one every few weeks.
最新的一篇其实就在昨天。
The most recent one is actually from yesterday.
我们从中能学到什么？
What do we learn here?
小心你许下的愿望。
Be careful what you wish for.
我还没准备好面对这么多关注。
I was not ready for all this attention.
这几乎把我压垮了。
It almost broke me.
我差那么一点就把整个东西删掉了。
Like, I was that close to just deleting the whole thing.
我不再接朋友们的电话。
I stopped answering my friends.
我甚至不想再看手机了，因为上面全是源源不断的消息。
I didn't even want to look at my phone anymore because it was just a stream.
而且，我的电话号码当然被人泄露了，还有很多私人信息，因为我是人类的敌人。
And of course, somebody leaked my phone number and also a lot of private details because I'm an enemy of humanity.
而我本来只是想做出点酷的东西。
And here I am just wanting to build something cool.
现在我们进入第二个问题。
And now we're getting to question 2.
你被收买了吗？
Did you sell out?
其实，这也不是我第一次经历这种事。
You know, this wasn't my first ride.
我二十多岁到三十多岁的时间都在做一家 B2B 软件公司。
I spent my, my 20s and my 30s building a B2B software company.
我像个原始人一样手写了一个 PDF 框架。
I wrote a PDF framework by hand like an animal.
我白手起家创立了公司。
I bootstrapped the company.
一路把它发展到了将近 80 个人。
I grew it to almost 80 people.
我完全无视了竞争。
I ignored the competition.
它成了大多数企业都会买的产品，最后我把业务交接给了联合创始人。
It became the thing most enterprises would buy, and eventually I passed things on to my co-founder.
我卖掉了股份，然后严重 burnout（倦怠）了。
I sold my shares and I very badly burned out.
退休。
Retirement.
这个想法听起来很不错。
The idea of it sounds great.
我花了将近三年时间漫无目的地游荡，弥补生活，参加各种聚会。
And I spent almost 3 years wandering around aimlessly, catching up on life and parties.
我做了所有以前生活中错过的事情。
I did all the things that I kind of missed out on in life.
有好几个月我连电脑都没开，就只是像普通人一样用手机随便刷刷网页。
I had months where I didn't even open my computer, you know, just casually checking the internet with my phone like a normie.
我一直知道，只要给自己足够的时间，那种冲动会回来的。
I always knew that if I give myself enough time, the urge would come back.
我一直以为冲动会是写代码。
I always thought the urge would be code.
我一直热爱编程，但花了一年时间才明白这么说并不完全准确。
I always loved programming, but it took me a year to understand that that's not entirely correct.
我热爱的是创造。
I love building.
编程只是实现目标的手段而已。
Programming was just a means to an end.
结果在我重新找回激情八个月后，收件箱里全是风投求着让我拿他们的钱。
And here I was, 8 months after I found my spark again, and my inbox was full of VCs begging me to take their money.
我自己都不知道想不想拿。
And I didn't know if I wanted to.
所以当那些大实验室找上门来，我突然跟 Mark、Sam 还有几个人通上电话的时候，
So when the big labs came knocking and suddenly I was on the phone with Mark, Sam, a few others.
这条道路看起来有意思多了。
This seemed like a much more interesting path.
而且感觉很不真实，对吧？
It also felt unreal, right?
别提什么冒名顶替综合征了。
Talk about imposter syndrome.
也许这对你来说是个重要的教训。
Maybe that's an important lesson for you.
你建造的一切——你建造的任何东西都可以被复刻或者克隆，但你的名字不行。
Everything that— everything you can build can be forked or cloned, but your name cannot.
所以你的个人品牌比你将来制造的任何单一产品都重要得多。
So your personal brand is way more important than any single product that you will ever build.
在你需要它之前就开始打造它。
Start working on that before you need it.
没错，我知道怎么玩这个游戏。
So yeah, I know how to play the game.
我也知道这么做等同于出卖灵魂。
I also know that doing this was the equivalent of selling my soul.
懂的人都懂。
If you know, you know.
这个决定并不容易。
The decision was not easy.
但如果我在 burnout 之后那段时间学到了什么，那就是这个。
But if I learned anything during my time after the burnout, it's this.
相信你的直觉。
It's trusting your gut.
而我的直觉告诉我，我最喜欢 OpenAI。
And my gut told me that I like OpenAI the most.
Hermes 赢了吗？
Did Hermes win?
我能给你的简短回答是，他们在要命的地方打败了我们。
And the short answer I can give you is they beat us where it hurt.
我们还在那里，而这两件事都是我的错，也都有我的份。
We're still there, and both of those are my fault and my doing.
发布后的几个月里，我们被安全报告彻底压垮了。
In the months after the release, we got absolutely crushed by security reports.
在某种程度上，我们就是现在许多开源项目所经历情况的雏形。
In some ways, we were the prototype of what many open source projects now experience.
尽管大多数都是非常边缘的案例，我还是感到了压力。
And even though most were super edge case, I felt the pressure.
然后就是媒体。
And then the press.
媒体说我们 20% 的技能是恶意的。
The press said 20% of our skills are malicious.
实际上我们专门写了一篇论文谈这个。
Now, we actually wrote a paper about this.
我们算了数据，实际数字更接近 0.3%。
We did the numbers, and the numbers are more like 0.3%.
我们把全部六万七千个都扫了一遍。
We scanned all 67,000.
论文我们发了。
We shipped the paper.
但你知道，勘误传播得永远没恐慌那么远。
But you know, a correction never travels as far as a scare.
我感到了这份责任。
I felt this responsibility.
世界发现了 OpenCLAW。
The world discovered OpenCLAW.
而且，虽然有一个巨大的吓人免责声明——实际上比我塞进幻灯片的大得多——但我知道很多人安装的时候不会去读文档，所以我真的专注在这一点上：加固代码库，构建一层又一层的安全机制。
And even though there was this big scary disclaimer, actually much bigger than I could fit on the slides, when you install it, I knew many people wouldn't read the docs, so I really focused on that: hardening the code base, building layers and layers of security.
我做了沙箱、白名单、还内置了权限的 Web 协议。
I did sandboxing, allow lists, a web protocol with permissions built in.
有些文件操作我们调用了 Python，因为 TypeScript 里缺少一些基本功能来确保你的 agent——
We were shelling out to Python for some file operations because there were some primitives that are simply lacking in TypeScript to ensure your agent.
——待在工作区里，不跟随符号链接，配置文件是原子写入的。
Stays in the workspace, it doesn't follow symlinks, and that the configuration files are written atomically.
大部分用户压根不在乎这些。
Most of the users didn't care shit about that.
当然，他们嘴上喜欢安全这种抽象概念，但实际上，我一更新就破坏了他们依赖的东西，降低了速度，还让更新变得更麻烦。
Sure, they, they liked the abstract term of security, but in all practical terms, they updated, I broke something they depended on, I made things slower, and I made things more difficult to update.
还有件事我得承认，我变懒了。
And another thing I have to admit, I got sloppy.
因为我现在得——我的时间要分给开源、媒体、花无数时间跟律师打电话去设立一个美国的 501(c)(3) 非营利组织。
Because now I had to— my time shifted between open source, the press, spending countless times on the phone with lawyers to set up a 501(c)(3) American nonprofit.
我还要办 O-1 签证，而且你知道，OpenAI 本身就是一个既有趣又要求很高的世界。
I worked on my O-1, and you know, OpenAI is its own world that is interesting and demanding.
我找了帮手，我们有了一个非常棒的、备受尊敬的维护者社区。
I brought on help and we got such an amazing community of maintainers that are very much respected.
每个人都加了点自己的小功能。
And everyone added their own little feature.
我觉得自己有点骑虎难下。
I felt like I'm a bit in a hard place.
毕竟这些人都是免费干活，对吧？
After all, these people work for free, right?
那我凭什么告诉他们该做什么？
So who am I to tell them what to do?
而且我的精力也分散了。
And also my attention was all over.
所以我们加了很多功能。
So, we added a lot of features.
你知道，功能才是好玩的部分。
You know, features are the fun part.
一个新功能就是一个提示词的事儿。
A new feature is just a prompt away.
真正的代价在后面。
The real cost comes after.
我们发布的每个功能当然都带一个配置选项，因为我们不想破坏每个人的设置。
Every feature we shipped, of course, with a configuration option because we didn't want to break everyone's setup.
在峰值的时候——我得数一下——如果把所有排列组合算进去，我们最后大概有九千五百个配置选项。
At our highest— I had to count that— we ended up with around 9,500 configuration options if you count all the permutations.
你可以写再多测试，也不可能覆盖所有这些配置，时不时就会出问题。
You can write all the tests you want, it is impossible to cover all of these and not break things from time to time.
有用户的软件，进化起来要难得多得多。
It is infinitely harder to evolve software that has users.
与此同时，其他公司有风投资金驱动，他们一路猛进。
Meanwhile, other companies, they were fueled by VC money, they pushed ahead.
那个动漫女孩公司就做得特别好。
The anime girl company did it especially well.
人们开始在任何Twitter对话中插一脚，而我们当时正埋头于安全基础和功能开发。
People started inserting themselves into pretty much any conversation on Twitter while we were buried in security groundwork and features.
他们有一个更简单的故事、激进的营销活动，还有一句吸引用户迁移的口号。
They had a simpler story, an aggressive marketing campaign, and a one-liner to migrate people's claws.
不过，最让项目受挫的还是Anthropic。
The main thing though that really hurt the project was Anthropic.
不是名字的问题。
Not the name.
那部分确实让人压力很大，但我多少能理解，而且他们人挺好的。
That part was stressful, but I kind of understood, and they were really nice about it.
问题在于我过度优化了他们的模型。
It was the fact that I optimized too much on their model.
我用Codex和GPT搭建了OpenClaw，但很长一段时间里，那个框架真正最好用、且针对Opus做了优化。
I built OpenClaw with Codex and GPT, but the harness for a long time really worked the best and was optimized for Opus.
所以当他们提前大概24小时通知我说要取消所有人的订阅时，
So when they pinged me with around 24 hours notice that they're going to disable the subscription for everyone.
真的没有足够的时间来调整方向。
There was not really enough time to change course.
我的意思是，我们当然支持开放权重模型。
I mean, sure, we support open-weight models.
我们在那上面做了很多工作，但它们当时确实还不算太好。
We did a lot of work on them, but they really weren't that great yet.
而且早期的OpenAI模型就是缺乏个性。
And the early OpenAI models simply lacked character.
所以这句话你最好记下来：
So maybe write this one down.
你所依赖的公司的商业模式就是你的商业模式。
Your dependency's business model is your business model.
现在这个问题已经解决了。
You know, that's all fixed now.
所以现在感觉挺棒的。
So it's pretty awesome.
开放权重模型现在确实很好用了。
Open-weight models are actually good.
我在框架工程上学到了很多，但很多人已经转向了别的东西。
I learned a lot about harness engineering, but in many ways people moved on.
从我们的下载数据就能看出来，五月份每周下载量跌到了八十三万五千次，然后六月份被宣布死亡之后，又冲到了四百七十万，创了历史新高。
You can see that in our download charts, we bottomed out at around 835,000 weekly downloads in May, and then after being declared dead in June, we peaked at 4.7 million, the highest ever.
这两件事同时都是真的。
Both of those are true at the same time.
炒作就像天气一样。
Hype is like the weather.
你也许能预见它，但控制不了。
You might see it coming, but you can't control it.
就我而言，那是一场风暴。
In my case, it was a storm.
问题四：做这件事还有意思吗？
Question 4, is it still fun?
大概在二月份的时候，它不再有趣了。
You know, somewhere around February, it stopped being fun.
我开始觉得——这变成了一种责任。
I started to feel like— it started to feel like a responsibility.
我醒来发现，一个原本不想再建公司的人，却落到了有两份工的地步。
I woke up and the man who didn't want to build another company found himself in a situation where he had 2 jobs.
或者应该说是一份工作和一个使命？
Or should I say a job and a calling?
我当时很挣扎。
I was struggling.
最糟糕的是，我停止使用自己的产品了。
And worst of all, I stopped using my own product.
差不多在那段时间，我不再做我自己喜欢的产品，而是开始做适合所有人的东西。
Somewhere in that time, I stopped making a product I love, and I worked on making something for everyone.
它不再是我每天都会用的东西，反而变成了一种我看到就感觉像是工作的东西。
It became less of a thing that I use every day and more a thing that I saw and felt was work.
我想象一下，“所有人”到底都是谁。
And I wanted to picture who everyone really is.
其中一个人有个买菜助手。
One of them had a grocery agent.
另一个人试图用社交工程手段攻击我的机器人。
Another one tried to socially engineer my bot.
在所有人添加功能和各种组织工作之间，我变成了那个修bug、修安全问题、提供支持、打基础的人。
Between everyone adding features and all the organization work, I became the person that fixed the bugs, fixed the security issues, and provided support and built a foundation.
而且因为大家都在传谣言，说什么OpenClaus是OpenAI旗下的，我就不太想从OpenAI那里拿太多帮助。
And because all these people were like giving the rumor mill, oh, OpenClaus owned by OpenAI, I didn't want to take too much help from OpenAI.
我是说，对，他们给了我一些token，而且我确实用得很猛。
I mean, yeah, they gave me tokens, and boy, did I use that.
但我还被卷入了其他项目，而且有一大堆人涌过来想跟我说话，我没办法轻松地拦住他们。
But I was also pulled into other projects, and there was a whole flood of people that I couldn't as easily block who wanted to talk to me.
事后想想，我本来可以很多事情做得不一样的。
In hindsight, I could have done many things differently.
比如说，我可以多要些帮助。
Like, I could have asked for more help.
我可以把责任从自己身上移开。
I could have moved responsibility off my plate.
但我陷得太深了，没花时间去想战略方向。
But I was so deep in everything that I didn't take time to think through strategically.
幸运的是，我遇到了一群很棒的人，慢慢地，事情就开始走上正轨了。
Luckily, I met a bunch of amazing people, and little by little, things aligned.
我搞定了签证，最后还成立了一个非营利组织。
I figured out my visa, eventually formed a nonprofit.
我们获得了很厉害的公司作为捐赠方，我也找到了一些真正相信开源的人，开始跟我一起工作。
We got amazing companies as donors, and I found some really good people that believe in open source and started working with me.
我还要特别感谢一下NVIDIA，因为他们很早就来帮忙，而且只是简单地问了我需要什么。
I also need to give NVIDIA, a special shout out because they were very early and they simply asked me what I need.
然后他们就派人接手了大部分安全工作。
And then they sent people to take over much of the security work.
我想大概是在五月份，我生日前后，我重新有了那种感觉——事情又开始变得美好了，那种建造东西的快乐又回来了。
And I think it was somewhere in May, around my birthday, where I had this feeling that things are starting to feel good again, where the joy of building for me was coming back.
而且是啊，媒体那边，隔一周就炮制出一个OpenClaus杀手。
And yeah, somewhere in the press, The press coined an OpenClaus killer every other week.
有一阵子，我觉得我数到二十个了。
At some point, I think I counted 20 of them.
甚至还真有一个项目叫OpenClaus杀手。
There's even literally a project called OpenClaus Killer.
是个卸载工具，这完全没必要，因为我们自己就有卸载工具。
It's an uninstaller, which is totally unnecessary because we have an uninstaller.
但是，你知道，这些杀手故事没有一个真正抓到重点。
But, you know, but none of the killer stories ever picked up what this is actually about.
重点是开源。
Open source.
我的灵感来源是被惹恼。
My source of inspiration is being annoyed.
现在，当我用到一个不能直接发条指令给我的Agent去改它的软件时，我就会觉得很烦。
These days I get annoyed when I have to use software where I can't just send a prompt to my agent to change it.
那就是那种乐趣回来了。
That's the fun part that's coming back.
很难跟一个纯粹在找乐子的人竞争。
It's hard to compete with someone who's just there having fun.
乐趣就是速度。
Fun is velocity.
我享受搭建的那些周，产品肉眼可见地变好了。
The weeks I enjoyed building The product got visibly better.
我不享受的那些周，我们发布了配置选项。
The weeks I didn't, we shipped config options.
问题五，接下来是什么？
Question 5, what's next?
图表！
Graphs!
对，有些观众可能会想，终于到了他讲图表的环节了吗？
Yeah, some people in the audience might be like, is this finally the part where he talks about graphs?
首先，我对我们基础层目前的状态非常满意。
First of all, I'm super happy with where we are with the foundation.
我们的使命是让人们更接近AI。
Our mission is to bring people closer to AI.
而且事情变化得太快了。
And things are changing so fast.
对很多人来说，这让人感到害怕。
For many people, it feels scary.
我对OpenCLAW取得的一项成就感到自豪。
I'm proud of one thing that OpenCLAW achieved.
那就是，对很多人来说，它把AI从那种模糊又可怕的东西，变成了有趣又奇怪的东西——你懂的，像龙虾什么的。
And that for many people, it moved AI from this thing that's like nebulous and scary into something that is fun and weird, you know, lobsters and all of that.
我们会继续朝这个方向努力。
And we'll keep pushing there.
继续构建一个伟大的开源软件生态系统，包括举办让人们相聚的活动，以及教育。
Keep building a great ecosystem of open source software with events that bring people together and with education.
我们目前有十名在册员工。
We have 10 people now on payroll.
我们还在招聘一些职位，包括一位CEO。
We're hiring a few more roles, including a CEO.
其次，我有点把claw变成了一个名词。
Second of all, I kind of made claw a noun.
你知道，Karpathy去掉了open。
You know, Karpathy dropped the open.
Satya说企业级claws。
Satya says enterprise-grade claws.
在微软的主题演讲中。
In Microsoft's keynote.
有三万三千个以Claw命名的代码仓库。
There are 33,000 Claw-named repositories.
如果我们身处一个模拟世界，那我们肯定是在一个比较奇葩、不会被关闭的模拟里。
If we are in a simulation, we are certainly in one of the weirder ones that won't get shut down.
我喜欢未来的这部分。
I dig this part of the future.
第三，我们仍然没有一个永远在线、永远思考的智能体。
Third, we still don't have an agent that's always on, always thinking.
还有太多事情要做。
There's still so much to do.
AI领域的格局和技术演进速度，比我们围绕它构建的软件还要快。
The landscape and the tech in AI are evolving faster than the software that we build around.
这对于你们所有人来说都是一个机会。
And that's an opportunity for all of you.
我们的工作流程也在演变。
Our workflows are also evolving.
就像我早期的一个愿景：我们不应该考虑会话或压缩。
Like the vision I had early on where we shouldn't think about session or compaction.
世界终于——模型和技术终于达到了一个阶段，让这成为现实。
The world is finally— the models, the tech is finally getting to a stage where this is becoming a reality.
我们正在进入一个世界，终于从文本、纯文本界面转向语音和多模态。
We're entering a world where we finally move on from the text, text-only interfaces to voice and multimodality.
就在昨天，我们搞出了一个hack，现在你的Claw可以给你打FaceTime了。
Just yesterday we got a hack working that your Claw can now FaceTime you.
而这正是OpenClaw存在的意义所在。
And this is kind of what matters why OpenClaw exists at all.
每个实验室都会卖给你一个agent。
Every lab will sell you an agent.
OpenClaw就是那个替代方案。
OpenClaw is the alternative.
开源，到处都能跑，支持任何模型。
Open source, runs everywhere, works with any model.
如果你用本地模型，你的数据就永远不需要离开你的设备、你的agent、你的电脑、你的生活。
And if you run local models, your data never has to leave your device, your agent, your machine, your life.
这甚至都不是我的论点。
That's not even my thesis.
那是Gary的。
That's Gary's.
你知道，我们只是先把它做出来了。
You know, we just shipped it first.
所以我们也终于再次用OpenClaw来构建OpenClaw了，不过这次有个新花样——每个人都能在团队服务器上看到彼此的会话。
So we also finally are building OpenClaw with OpenClaw again, but this time with a twist where everyone sees each other's session on a team server.
而且有一个Claw知道每个人在做什么，还能接管工作的编排。
And there's a Claw that knows what everyone else is working on and can also take over orchestration of the work.
我们终于慢慢摆脱了这个未来世界里的奇怪现象——人们拿着打开的笔记本电脑到处跑，因为agent需要持续工作。
And we are finally leaving this slowly, this weird blip in the future where people work in terminals and run around with their laptops open because the agent needs to keep working.
如果你们只记住三件事的话。
If you only remember 3 things.
第一，不要停止找乐子。
Number 1, don't stop having fun.
乐趣是最终的驱动力，这样你才能获得最好的想法。
Fun is the ultimate driver so you can, you can get the best ideas.
第二，相信你的直觉。
Number 2, listen to your gut.
同时也要修复那些让你烦恼的东西。
And also fix the things that annoy you.
它可能就会成为下一个大热门。
It might just become the next big thing.
第三，保持专注。
And number 3, stay focused.
你知道，再多录一档播客并不会让你赢。
You know, another podcast is not what will make you win.
活在未来，构建缺失的东西。
Live in the future, build what's missing.
当人们为你写讣告时，继续交付。
And when they write your obituary, keep shipping.
这会让他们搞不清状况。
It confuses them.
好了，刚才这是五个问题的回答。
Now, these were 5 questions.
我相信你们还有更多问题。
I'm sure you have many more.
而且我想我们还有充足的时间进行问答。
And I think we have plenty of time for Q&A.
所以先是agents。
So first it was agents.
然后是loops，再然后是graphs。
Then loops, then graphs.
你们现在是怎么搭建的？
How are you building these days?
我的意思是，一直都是在session里，对吧？
I mean, it was always sessions, right?
但是一开始，你真的得注意清理，还要确保你的指令是连贯的。
But in the beginning, you had to actually care that you would clean and you would make sure that your instructions are coherent.
而现在，我的会话更像是主题。
And these days, it's more that my sessions are topics.
有时候清空会话甚至是个劣势，因为里面有大量信息能帮助智能体。
Clearing the session sometimes is even a disadvantage because there's so much information in there that helps the agent.
但最大的、最大的、最大的变化是，我试着让智能体更主动地为我工作。
But the big thing, the big thing, the big thing that shifted is I try to make the agent do more proactive work for me.
所以当我把注意力转移到某件事上时，我不想看问题列表。
So when I shift my attention to something, I don't want to read issues.
我想看到已经经过完整审查和测试的PR。
I want to see fully reviewed and tested PRs.
也许我喜欢这个功能，也许我不喜欢，但我不想分散注意力。
Maybe I like the feature, maybe I don't, but I don't want to split my attention.
同时在工作中，如果你来找我，跟我说这个功能想法，我会生你的气。
At the same time at work, if you come to me and you tell me this feature idea, I'm going to get mad at you.
这太容易了：你只需要和智能体讨论功能想法，把它做出来，截图，然后让我玩一玩。
It's so easy that you just discuss the feature idea with an agent, you build it, you make screenshots, you let me play with it.
这样，如果它不错，我们可以立刻迭代。
That way, if it's good, we can immediately iterate on it.
但大多数时候，人们会自己发现为什么不好，甚至根本不会来找我。
But most of the time, people will figure out why it's not good and don't even come to me.
图之后是什么？
What comes after graphs?
而OpenClaw下一步是什么？
And what's next for OpenClaw?
我知道，有时候我喜欢在周日发点垃圾帖子，然后我的Twitter就炸了。
I know, sometimes I like to shitpost on a Sunday and then my Twitter explodes.
但这也未必是垃圾帖子，对吧？
But also it's not a shitpost, right?
那么任何时候你——你知道我们作为工程师在很长一段时间里做了什么吗？
So anytime you— you know what we as engineers did for the longest time?
我们构建自动化来让生活更轻松。
We built automations to make our life easier.
自从我们这个职业存在以来，就一直如此。
That's been going on since our profession exists.
所以如果你设计，无论你叫它循环、图还是工作流，其实都差不多。
So if you design, call it loop, call it graph, call it workflow, it's kind of all the same.
如果你设计一个东西，它收到触发、得到输入，然后为你做点什么。
If you design something that gets a trigger, gets an input and does something for you.
也许中间还有一个决策，砰，你的图就出来了。
And maybe there's a decision in the way, boom, there's your graph.
这并不神奇，尽管我收藏了一些关于图工程的文章，因为我有点好奇他们的观点。
It's not magical, even though I favorited some of those graph engineering posts because I'm kind of curious what their opinion is.
确实如此。
That it would be.
在我的世界里，这不过是我们早已做了很久的自动化故事的一个更好的方式。
In my world, it's just a better way of our automation story that we did for so long already.
在一个快速构建、快速交付的世界里，如何确保你所构建的东西仍然可靠且可扩展？
In a world of build fast, ship fast, how to make sure that what you build is still reliable and scalable?
是啊，这部分我曾经搞砸过一段时间。
Yeah, that's the part I messed up for a while.
因为我不够专注，而且模型在测试方面也不太好。
Because I was not focused and the models were not really good at testing.
我认为现在的模型已经非常出色了。
I think that the models are really good now.
我们不仅有会记忆的会话，我们还把编排能力训练进了模型。
We have, we not just have sessions that remember, we have orchestration trained into models.
所以它们真的理解并知道如何使用子智能体。
So they really understand and know to use subagents.
我们有计算机使用功能。
We have computer use.
我们有浏览器使用功能。
We have browser use.
所有这些加在一起，就像一个完美的问答环境。
Like all of that together is like your perfect Q&A environment.
比如我昨天就这样做了，启动了 Codex，用了12个子代理，理解我的项目，把它分解成功能。
Like I did that yesterday where I spin up Codex, use 12 subagents, understand my project, break it down into features.
然后每个子代理会进行压力测试或代码审查，然后告诉其他会话应该重点测试哪里。
And then each subagent would like stress test a feature or code review a feature and then inform the other session where to focus testing on.
我们还没有到可以自动化一切的地步。
We're not yet at the point where you can automate everything.
你还是需要做一些手动点击，感受一下效果。
You'll still need to do some manual click-throughs so that you know how it feels.
但对于很多用户会遇到的典型 bug，你现在其实可以通过提示词来处理，而且效果很好。
But like for a lot of the typical bugs that users will encounter, you can actually now prompt your way and you get very far.
好的。
Okay.
AI 工具。
AI tools.
你纯粹为了速度而做的决定是什么？
What's your decision you made purely for speed?
早期做的，到现在还在用的？
Early on that you still live with today?
我觉得我很早就决定不读所有代码。
I think I was really early in deciding that I don't read all the code.
我把代码审查更多地看作风险管理。
I see code review more as risk management.
有时候你碰到一个棘手的系统，就会想仔细读一下，有时候你在做 UI。
Sometimes you touch a system that is scary, kind of want to read a little bit closer, and other times you build UI.
如果 UI 看起来正确，我真的在乎它是怎么构建的吗？
Do I really care how the UI is built if it looks correct?
并不。
No.
所以你大概扫一眼，或者直接接受它看起来没问题。
So you kind of glance over or you simply accept that it looks right.
一部分原因是你慢慢培养出一种感觉，知道某件事大概需要多长时间。
Part of it is just like you develop a little bit of a feeling how long something should take.
所以如果我做一个小的调整，应该改变拖拽的工作方式，却花了3个小时，我就知道有问题了。
So if I do a small tweak that should change how dragging works and it takes 3 hours, I know something's wrong.
我会仔细看。
I'll look closely.
但除此之外，代码审查的一部分就是观察，看改动有多大，然后相信直觉。
But otherwise, part of code review is simply like observing, looking how big a change is, and trusting your gut.
好的。
Okay.
AI 工具让现在一个周末就能做出一个产品变得容易。
AI tools make it easy to build a product in a weekend now.
但构建并不是最难的部分。
But building isn't the hard part.
让人们使用它才是。
Getting people to use it is.
现在你会如何从一个能用的原型发展到最初的10个真实用户？
How would you go from a working prototype to your first 10 real users today?
嗯，我认为那是我在演讲中提到的一部分。
Well, I think that was part of what I addressed in my talk.
第一个用户应该是你自己。
User number 1 should be you.
而我的第2到第20个用户是朋友。
And my users 2 to 20 were friends.
所以我肯定你有一些人会实际测试你的东西，但你自己需要是第一个用户。
So I'm sure you have some people that would actually test your stuff, but you need to be the first user.
如果你对自己正在构建的东西不感到兴奋，那可能就没意义了。
If you don't get excited about what you're building, it'll probably not make sense.
比如，在当今这个时代，眼球几乎是最昂贵的货币，因为现在开发东西太快了。
Like, in this day and age, eyeballs are kind of the most expensive currency because building something so fast now.
你如何平衡解决那些让你烦恼的问题和开发你认为人们想要的功能？
How do you balance solving issues that annoy you and make features you think people want?
这是个难题，因为很多时候它们是相辅相成的。
That's a hard one because oftentimes it kind of goes together.
如果有些东西不符合我的期望，我就会感到烦恼。
I get annoyed if something's not there as I want it.
老实说，我还没有完全想明白那部分。
I haven't, to be honest, I haven't fully figured that part out yet.
我觉得我可能会在问题上花上好几个月，因为软件中总有奇怪的边界情况，尤其是当你添加越来越多的功能时。
I feel I could just work for months on issues because there's always going to be weird edge cases in software, especially if you add more and more features.
但话说回来，如果我光做那件事，我就会对工作失去兴趣。
But then again, if I only do that, I'll lose interest in working.
所以它需要是一种健康的混合。
So it needs to be like, it needs to be a healthy mix.
如果你能回到过去，改变关于 OpenClaw 的一件事，你会改变什么？
If you could go back and change one thing about OpenClaw, what would you change?
我会对安全研究人员不那么紧张。
I would be less stressed out about security researchers.
他们很擅长让你感觉很糟糕。
They are really good at making you feel really bad.
他们会发送报告，给我发邮件，给我打电话，尽一切可能引起我的注意，但实际上并不是为了帮助产品。
They would send a report, they would email me, they would call me, they would do everything they possibly can to get my attention, but not actually to help the product.
在大多数情况下，他们只是为了获得影响力，为了得到一点成就。
In most of the cases, it's really just for them to get clout, for them to get a point.
“哦，我们发现了什么。”
Oh, we found something.
而且他们中的大多数人发送的报告都是他们的代理生成的，甚至没有实际测试过。
And most of them really sent reports that their agent produced without actually even testing it.
我会采取更强硬的立场，解释哪些部分是我们保证的，哪些部分不会修复，因为那不是我们的安全边界。
I would take a stronger stance of explaining what are the parts that we guarantee and what are the parts that will not be fixed because that's not our security boundary.
但老实说，我当时完全不知道。
But honestly, I had no clue.
这是我第一次接触这个世界。
This was my first time I was exposed to this world.
所以我就是不知道。
So I just didn't know.
该怎么处理。
How to handle that.
我浪费了几个月，还长了白发，才学会。
And I lost a few months and got a few gray hairs to learn.
当今代理基础设施中最大的单一瓶颈是什么？
What's the single biggest bottleneck in today's agentic infrastructure?
可靠性、工具、内存、评估、计算管理，如果这样说合理的话。
Reliability, tooling, memory, evals, managing compute, if that makes sense.
如果我在本地机器上运行一个测试，TSGO 会启动 16 个线程，然后让我的机器成为瓶颈。
If I run one test locally in my machine, TSGO will spin up 16 threads and will bottleneck my machine.
如果 10 个会话同时这样做，其中 2 个可能会超时，我们就得重来。
If 10 sessions do that, 2 of them will probably time out and we'll have to do it again.
而且没有一个真正好的系统来管理所有这些。
And there's no really good system of managing all of that.
现在，如果我做网页相关的工作，很容易创建一个云端会话。
Now, if I do web stuff, it's easy to create a cloud session.
如果我要做需要 macOS 的事情，那已经是 99% 的工具所在的地方了。
If I do something that requires macOS, that's already where 99% of the tools
如果它需要我电脑上的其他东西，我也是。
If it's something that needs other stuff that's on my computer, same.
而且我们还没有真正构建出一个东西，能让东西毫无问题地轻松从这里移动到那里。
And we haven't really built something yet where things could easily move from here to here without issues.
至少我还没有看到什么好用的东西。
At least I didn't see good stuff working yet.
或者能让我可靠地管理一群设备。
Or where I could drive a fleet in a reliable way.
就比如现在，我用了太多系统。
Like right now, I use too many systems.
我通过屏幕共享进入一些电脑来分散负载。
I screen share into some computers to distribute the load.
我不应该那样做。
I shouldn't be doing that.
你如何让一个开源项目保持自己的主张和明确的方向，当有人只是提交了一个第N+1个PR来添加一个随机的功能时，你怎么办？
How do you keep an open source project opinionated and pointed in one direction, whether they're just merging the N+1 PR that adds a random feature?
你又如何向社区传达这个方向？
And how do you communicate that direction to the community?
你是否曾经拒绝过一个很受欢迎的PR，因为它让项目偏离了方向？
Have you ever said no to a popular PR because it pulled the project off course?
是的，是的。
Yeah, yeah.
我得说我拒绝得还不够多。
I would say I didn't say no enough.
现在，当我开始一个新的开源项目时，我会写一个vision.md文件，在里面大致解释一下项目现在的状况，以及我认为它未来的发展方向。
These days, when I do new open source, I write a vision.md file where I kind of explain what it is now, where I see this to be.
但当然，这并不准确。
But of course, it's incorrect.
它并不完美。
It's not perfect.
完美的科学？不是的。
Perfect science.
而且我需要更好地遵守它，因为总是会有诱惑去添加一个看起来很酷的功能。
And it's something I need to adhere to better because it is always tempting to add this one feature that looks really cool.
但你很少想到的是，当你合并这个功能时，它实际上意味着一堆我不需要的代码，那个提交者可能并不真正理解，我也没完全理解，而且我还得为之负责。
But what you rarely think about is when you merge this feature, it really means here's this pile of code that I don't need, that the person probably doesn't really understand, that I don't fully understand, that I also got responsibility.
你预计什么时候能有那些持续运行的“爪子”？
When do you expect to have claws that are perpetually running?
那些主动工作的东西。
That are proactive in their work.
我的意思是，老实说，这更多不是技术问题。
I mean, honestly, that's not so much a tech problem.
更多的是token的问题——我们今天就能做到，但你的订阅可能撑不了多久。
It's more a token problem where we could do that today, but you wouldn't get very far with your subscription.
而且不是每个人都愿意花那么多token。
And not everyone is willing to spend so many tokens.
而且——但困难的部分在于设计一个不会白白浪费token的系统。
It's also— but the part that's not easy is designing a system that doesn't just burn empty tokens.
即使是我早期的那个心跳系统，也太静态了，不够主动。
Even my early system of like heartbeats was too static and not proactive enough.
如果你有一个很大的会话（session），那就尤其糟糕。
And then especially bad if you have a large session.
然后一小时后，你给所有东西搞一次心跳检查。
And then 1 hour later you call like a heartbeat checkup on everything.
那意味着在KV缓存清空后，你向服务器发送了60万个token，然后为很多没用的工作花了一大笔钱。
And that means you send 600,000 tokens back to the server after the KV cache cleared and you pay a stupid amount of money for a lot of not useful work.
所以有很多可以优化的地方。
So there's a lot of things that can be done to optimize that.
这也真的非常非常难。
It's also really, really hard.
你能再多讲讲你自己的个人设置吗？
Can you tell us more about your own personal setup?
你的代码托管在哪里？
What do you host your code on?
你用的是哪个模型？
What model are you running?
你用的是哪个工具框架？
What harness do you use?
你还会阅读——能不能滚动一下屏幕？
Do you still read— can we scroll?
你还会读自己的代码吗？
Do you still read your code?
我用的是 MacBook，但我通常用 Jump Desktop 远程屏幕共享到我的工作室。
I use my MacBook, but I usually use Jump Desktop to screen share into my studio.
然后就用那台电脑，因为那台一直在运行。
And then just use the computer there because that one's always running.
那台机器可以随便跑什么，速度也快，还不用消耗我的电池。
That one I can run whatever I want fast and not drain my battery.
而且我可以合上笔记本，事情还在继续跑。
And I can close my laptop and things keep working.
我还有几台远程机器，有时候会用 VNC 连过去。
And I have a few other remote machines where I sometimes VNC in.
这样做最棒的一点是，因为我做很多 Mac 软件，AI Agent 很喜欢接管我的屏幕到处点。
The part that's really nice about it is because I do a lot of Mac software, agents love to take over my screen and click around.
如果你给它们自己单独的机器，就不会打扰到你了。
And if you put that in their own, if you give them their own machine, it will not bother you.
不然的话，你就会跟 Agent 抢鼠标光标。
Otherwise, you'll fight with the agent for the mouse cursor.
你还会读自己的代码吗？
Do you still read your code?
我觉得我回答了一点，就是风险管理这个概念。
I think I answered that a little bit with the term risk management.
而且，如果是我自己做开源，风险管理和我在 OpenAI 做软件是不一样的，在 OpenAI 我们仍然会通读所有代码。
And also, if I do open source by myself, it's a different risk management than if I do software at OpenAI, where we do still read all of the code.
Peter，你会怎么启动你的下一个创业想法？
Peter, how would you approach your next startup idea?
我不知道，我该把这个告诉所有人吗？
I don't know, should I tell this to everyone?
我觉得说出来会被很多人抄。
Like, I feel like it gets a lot of copies.
我觉得最重要的是，你得做一个你自己想用的东西。
I mean, the most important thing is you need to build something that you want to use.
否则它就不会好。
Otherwise, it'll just not going to be good.
其次，我希望你在个人品牌和可见度上下了功夫。
Second of all is I hope you worked on your personal brand and on visibility.
因为在当今这个时代，噪声太多了，你最困难的问题不是技术，不是软件，甚至不是人才。
Because in this day and age, there's so much noise out there that your hardest problem is not the tech, it's not the software, not even the people.
现在最难的问题就是吸引眼球。
The hardest problem now is like getting eyeballs.
也许我会再选择一个既难又无聊的领域。
Maybe I would pick something again that's in the category hard and boring.
因为这类领域通常更容易找到真正欣赏你成果的人。
Because that's usually a category that is a little bit easier to actually find people that will appreciate.
当你解决了一些问题之后。
When you solved something.
如果你选一个虽然难但好玩的东西，你的处境会非常艰难，尤其是在现在人们随便用提示词就能生成东西的时代。
If you pick something that is fun, even if it's hard, you're gonna have a very tough time, especially in a time where people can just prompt things into existence.
好的，今天就到这里。
All right, last one for today.
你想要别人做什么产品？
What's a product you want someone to build?
你知道吗，我觉得弄一个 Linux 的测试机箱很容易，但要搞一个 Mac 上运行得特别好的测试机箱，那难得不合理。
You know, I feel it's very easy to have— to get a test box for Linux. It is unreasonably hard to get one for Mac that works really well.
而且我搞到的那些 Windows 用的东西也挺烦人的。
And all the stuff that I got for Windows is also quite annoying.
我还没找到一个真正好的供应商，能把所有这些做得又快又便宜。
I haven't found a really good provider yet that does all of that fast and cheap.
我不确定这是不是你想要的生意，因为开发工具本质上很难做，但这是我特别想要的东西。
I'm not sure if this is the business you want to be in because dev tools are inherently hard, but that's something I would love to have.
谢谢大家。
Thanks everyone.