20VC：Codex 对比 Claude Code 对比 Cursor：谁胜谁负？所有编程都会被自动化吗？我们还需要产品经理吗？AGI 的真正瓶颈是什么？代理的三个阶段及你需要了解的一切——对话 OpenAI Codex 负责人 Alex Embiricos

我们正努力让人们更容易尝试不同的东西。

And we're trying to make it even easier for people to try different things.

话虽如此，这些编码任务——当你让代理编写代码时——是非常封闭的。

Now that said, these coding tasks, right, where you're asking an agent to write some code, they're quite hermetic.

我所说的封闭，可以用电视节目来打个比方，就像单元剧一样。

And what I mean by this is maybe an analogy in TV would be like episodic, right?

比如，你可以进来，有一个开放的、任何代理都能读取的代理文件，还有任何代理都能使用的技能，你可以让代理编写代码，它会生成一个补丁并提交到 Git 中。

Like, you can come in, and you've got this like open ended, like agents file that any agent can read from, you've got these skills that any agent can use, and you can ask the agent to write some code, and it produces a patch and that patch goes into git.

因此，目前这个过程的两端都非常中立、厂商无关，切换起来非常容易。

So kind of like both ends of this are pretty neutral, vendor neutral, so very easy to move between for now.

当代理开始从事的不再是编写代码，而是更通用的工作——无论是针对软件工程师还是其他构建者时，它们就需要开始与其他系统交互。

As agents start to do work that is not writing code, but more general work, again, for software engineers or beyond for any builder, they're going to need to start interfacing with other systems.

所以，当代理开始与 Sentry、Google 文档或其他系统通信时，我认为这些代理会变得更具粘性，因为决定让代理连接到某个系统是一个难以逆转的决策。

So as they start, maybe your agent is talking to Sentry, right, or it's talking to your Google Docs or something, then I think these agents become much stickier because actually deciding to connect an agent to that system is a sticky decision.

如果你是企业用户，真正信任代理能访问这些工具，那么确保有完善的安全防护、沙箱环境以及对代理如何与这些系统交互的控制机制，就显得至关重要。

And if you're an enterprise, really trusting that the agent is going to have access to these tools, but there are really good secure guardrails and sandbox and like controls over how the agent works with these systems, think is critically important.

这可不是你愿意反复去做的事情。

And that's not something that you're going to want to to do multiple times.

因此，我们在开发Codex时就已经预见到这一点。

And so, you know, we've been kind of building Codex knowing that this is coming.

所以我们采用了最保守的沙箱策略。

And so we have like the most conservative sandboxing approach.

沙箱本质上是一组操作系统级别的控制机制，用于限制代理的行为。

Sandboxing is kind of like a set of controls, OS level controls over what the agent can do.

但我很喜欢《七种力量》这本书，它深入探讨了企业获取价值和可持续性的七种方式。

But I'm a fan of Seven Powers, this brilliant book, which talks about kind of seven ways that businesses accrue value and sustainability.

而你的用户粘性或留存率就是其中之一。

And like your stickiness or your retention is one.

如果我们和柯达在同一团队，该如何建立持久的模式、行为和项目，确保人们留在Codex，而不是在有更好的模型时转向Cursor或ClawCode？

If we're on the same team with Kodak, how do we create retentive patterns, behaviors, programs to ensure that people stay with Codex and they don't flip to cursor when there's a better model or ClawCode when there's a better model?

是的。

Yeah.

我的意思是，这很有趣，因为一方面，我们当然在经营一家公司，但我们的使命是确保安全地将AGI的好处惠及全人类。

I mean, it's interesting because I think on the one hand, like we think about this, obviously, we're running a business, but, you know, our our mission here is to, like, ensure that, like, we safely deliver the benefits of AGI to all humanity.

所以，关于Codex团队，有些事情对人们来说是反直觉的。

And so something that's, like, unintuitive to people about, like, the Codex team.

亚历克斯，我其实我是我

Alex, you actually I'm I'm I'm

我知道，但你的职责是Codex的成功。

I know but your job is the success of Codex.

我明白。

I get

实际上，我们的职责是智能的分发。

actually, our job is the distribution of intelligence.

因此，我们显然在构建Codex，这对许多听众来说非常反直觉。

And so we're obviously building out Codex, and this is really unintuitive to a lot of listeners.

但我们投入了大量精力训练这些模型，然后却把这些模型提供给我们的竞争对手。

But, like, we put all this effort into training these models, and then we serve these models to our competitors.

从我们的角度来看，

And from our perspective,

作为一位风险投资家，这对我来说太难理解了。

this is so difficult for me as a venture capitalist to understand.

你对此是知情的。

You are aware of this.

是的。

Yeah.

我完全清楚这一点。

Totally aware of it.

就像我们OpenAI是一个非常有趣且非同寻常的工作场所。

It's like we're OpenAI is like a really interesting and unusual place to work.

但基本上，因为我们着眼于长远，如果竞争对手变得更强，我们就能学到东西。

But basically, because we're playing such a long game for us, if the competition gets better, we learn.

这实际上对我们有帮助。

It's actually helpful for us.

所以我们正在大力推动Codex的发展。

And so we're pushing really hard at growing Codex.

你们会学习吗？

Do you learn?

因为如果他们封闭起来并取得进步，你们就学不到东西了。

Because if if if they're closed and they improve, you don't learn.

我不这么认为。

I don't think so.

例如，最近有一些新产品发布。

For example, there are a bunch of recent launches.

比如今天早上，我刚刚转发了一条关于Warp发布的产品推文。

Like, today, I literally just, like, quote tweeted a thing this morning about a launch from Warp.

并没有任何特定的关联。

No particular affiliation.

对吧？

Right?

里面还有很多很棒的想法，关于他们如何设计代理，使其能够同时在云端和本地运行。

And there are a bunch of cool ideas in there about how they, like, framed up the way that their agent can work in the cloud at the same time as working locally.

对我来说，这非常有启发性。

And for me, that's like inspiring.

我看到了来自不同公司的各种这些成果。

And I think I see all these things from various companies.

我觉得这个领域最酷的一点是，我们所有人最终都会不约而同地得出相同的结论，然后各自去实现它们。

And like, one of the coolest things about the space is it's like, we're all kind of inevitably reaching the same conclusions together and then building things out.

所以，在Codex团队里，我认为我们有一些巨大的优势，对吧？

And so, you know, on the Codex team, think I we have some massive advantages, right?

我们拥有ChatGPT带来的巨大分发优势，拥有训练我们自己的模型并使其在我们的系统中表现优异的能力优势，同时构建出适配新模型的优秀系统，而其他人根本没有机会提前接触到这些。

We have the massive distribution advantage with ChatGPT, we have the massive like capability advantage of training our own models to be good in our harness, and building our harness to be good at the new models, and like no one else has early access to those.

因此，我认为我们是在为胜利而战，拥有一个或多个巨大的优势，但同时我们也在打一场持久战——因为我们把模型提供给所有人，推动开放标准，让每个人都能使用我们所倡导的一切。

And so I think we're we're playing to win and we have a a really big advantage or a number of advantages, but we're also playing this long game where, you know, again, we serve our models to everyone, where we push for open standards so that everyone can use, like all the things that we're pushing for as well.

我能问你一个问题吗？赢得胜利的决定性因素会是什么？

Can I ask you, what would be the defining factor of winning?

我知道我用了风险投资的术语，而你非常出色，更自由和开放。

And I know I'm using venture language and you're brilliant and kind of much more free and open.

但赢得胜利的关键因素是什么？我再问你一次，是市场推广策略吗？就是全球最大的企业是否愿意与OpenAI合作？

But it was like the defining factor of winning, again, I push you, is it like GTM, is which like the biggest enterprise in the world do want to work with OpenAI?

我有很多朋友在你们的销售团队里。

I have many friends in your sales team.

你们从最大品牌那里获得的主动咨询量简直不可思议。

The inbound that you get from the largest brands is incredible.

所以，是凭借卓越的品牌、产品执行力，以及Codex本身是个超棒的产品，还是计算与推理速度、真正的计算优势？

So GTM because of the incredible brand, product execution and just Codex being a fricking awesome product, or compute inference, speed, actual compute advantage?

哪一个才是决定性的胜利因素？

Which one is the defining winner?

好吧。

Okay.

所以，如果我们从OpenAI的角度来讨论这个问题，显然这超出了我的职责范围，但我会说，是计算优势和拥有最好的模型。

So I think if we're going to talk about it more from an OpenAI perspective, obviously, this is way above my pay grade, But I would say it's compute advantage and having the best models.

为了实现这一点，我们需要建立能够产生收入的业务。

And in order to achieve that, we then need to build businesses to generate revenue.

而且我们注意到，有了Codex团队——这个融合了研究和产品职能的团队——通过打造这些成功的产品，我们反而催生了更快改进模型的巨大压力。

And also that something we've that's really interesting we noticed with having the Codex team, which is a sort of combined team of research and product is also by building these these successful products, we create a lot of pressure to improve the model in sort of a faster way.

这可能是从公司角度来说的。

That's maybe the company perspective.

对吧？

Right?

如果我们从产品角度出发，我认为我们能做的最重要的一件事，就是打造一个人们真正想要使用的优秀产品。

If we come to the product perspective, I think the single most important thing we can do is build a a really good product that people want to use.

就像我之前说的，我们真正希望为个人打造产品，让人们对这些产品变得熟练，然后自然地融入自动化功能。

And like I was saying earlier, I think we really want to build products for individuals, and then allow, like, people to become fluent in those products, and then like, pull in automation.

我认为这可能看起来反直觉，但相比单纯从企业工作流角度出发，这种方式会带来更大的影响力。

And I think that may be counterintuitive, but will result in way more impact than anyone purely approaching it from like the enterprise workflow perspective.

我觉得这主要是个产品执行的问题。

You know, I think that's mostly a question of product execution.

这对专业消费者来说是有效的。

And then that works for say, like prosumer.

谈到企业市场，推广策略非常重要。

When it comes to enterprise, the go to market side is really important.

我通过惨痛教训学到的是，如果我们去接触企业，只是说：嘿，我们在这儿，随便用我们的东西，这是行不通的。

Something that I've learned the hard way is if we go to an enterprise, and we're just like, hey, we're here, like, feel free to use the stuff that doesn't work.

实际上需要做大量的教育工作。

There's actually quite a lot of education that needs to be done.

而且我们需要支持很多配置，并对整个团队进行培训。

And there's a lot of, like, configuration that we need to support and sort of like education of the broader team.

所以这个过程更像是：先介入，进行推介，与开发者体验负责人或其他相关人士会面，了解他们希望团队如何运作，然后提供工具，帮助他们将这种运作机制推广到整个团队。

So like, that motion looks much more like coming in, pitching, meeting the head of developer experience or whatever, understanding how they want their team to operate, and then giving them tools to like propagate that mechanism of operating to the rest of the team.

你刚才提到了‘收入’这个词，这是衡量企业的一个指标。

You said the word revenue there, which is one metric to measure a business against.

当你思考自己的成功指标时，你会坐下来和萨姆、布拉德或者其他人说：嘿，我们正在优化的目标是什么。

When you think about like your metric of success, which you sit down with Sam or Brad or whoever it is and say, hey, this is what we're optimizing for.

你们用什么指标作为衡量进展的核心目标？

What is the metric that you use as the defining north star for your progression?

其实不是收入，最主要的是活跃用户。

It's actually not revenues, the primary, the primary is active users.

你们怎么

How do you

你们如何衡量活跃用户？是日活跃用户吗？

measure active users, daily active users?

是的。

Yeah.

所以我们衡量的是周活跃用户，也就是这个人有没有真正使用过我们的产品？

So we measure weekly active users and it's, know, did this person like actually do a turn in our product?

他们有没有发送过一个提示？

You know, did they send a prompt?

你觉得周活跃用户这个指标足够频繁吗？

Is weekly active a frequent enough metric do you think?

听起来不错。

Sounds nice.

但如果这真的取代了IDE，每日活跃用户不是更好吗？

But if this is actually replacing the IDE, is daily active not better?

我认为每日活跃用户很快就会更好。

I think daily active will be better soon.

我们只是恰好使用每周活跃用户。

We just happen to use weekly active.

这里这是一种标准。

It's like a standard here.

而且我认为在我们起步时，这样做是有道理的。

And I think as we were getting started, it made sense.

但我真的同意那里的批评意见。

But I I I actually agree with the the criticism there.

我们应该干脆直接用每日活跃用户。

It's like, we should probably just be a daily.

我的意思是，我们应当进入这样一个时代：无论你面对什么任务，第一反应都是找一个智能代理来帮忙。

Like, I think we we need to be getting to a world where for any given task that you have, your first instinct is to ask an agent to help.

对吧？

Right?

就像使用谷歌搜索一样，不管需要做什么，我只要打开那个文本框，就能被引导到正确的地方。

It's kinda like, you know, how, like, with Google search, it's just like, okay, anything I need to do, I just like go into this text box and I can get navigated to the right location.

然后有了ChatGPT。

Then you had ChatGPT.

对于任何我需要的信息，我都可以打开这个文本框，打出来，就能获得有用的答案。

It's like for any information I need, I can go into this text box, type it out and get information that helps me.

我认为今年我们将看到的下一阶段是：针对任何我需要完成的任务——而不仅仅是获取信息——我只要进入这个文本框或输入界面，就会有某种帮助发生，哪怕它只完成了任务的一小部分。

And I think the next phase that we'll see this year is like for any task I need to do, as opposed to just get information, I go to this text box or this input, and something happens that helps me, even if it's not the full task, even if it's only a small part of it.

你刚才提到聊天这件事，我又跑题了。

You said about kind of chat that again, I jump around.

抱歉，我妈妈得陪我在伦敦散步，她得应对我这种躁动不安、跳跃式思维的大脑。

Sorry, my brain My mother has to walk with me around London and she deals with this manic episodic brain.

但你刚才提到聊天和那个界面。

But you said about chat and the interface there.

我对此非常着迷，因为对于忙碌的人来说，这似乎是一种极其高效的输入方式。

I'm really fascinated by this because it is a seemingly incredibly efficient input function for busy humans.

但我最近和安德森的全科医生阿尼什·阿卡亚聊过，他说不不不，这东西是萨姆和马斯克创造的，只适合效率极高的人。

But I spoke to Anish Akaya, who's a GP at Andreessen, and he came out the other day, and he's like, no, no, no, this was created by Sam and Elon and it works for very efficient people.

但地球上大多数人还是想要基于浏览器的发现式交互界面。

But most of the planet want browser based discovery interactions UIs.

你认为聊天会成为下一代AI与人类交互中持久的界面吗？

Do you think that chat will be the enduring UI in the next wave of AI interaction with humanity?

简单回答是肯定的，但其实我觉得这里有两个方面。

The simple answer is yes, but actually I think there's two components here.

比如，如果我们只是想象一下未来，就像看一部科幻电影那样。

Like, if we if we just imagine the future, like, just like, let's think of some sci fi movie.

对吧？

Right?

那么，人工智能看起来是什么样子？

Like, what does AI look like?

我相信科幻作品能很好地预测未来应有的样子。

I I believe that sci fi is a really good predictor of what the simple the future should look like.

而且通常它都很简单，因为故事需要简洁，我认为简单往往才是正确的。

And usually, it's pretty simple because it's the story, and I think simple is usually right.

它应该就是一个实体，我可以以任何方式、谈论任何话题和它交流，对吧？

It's gonna be some just, like, entity that I can talk to however I want about whatever I want, right?

我觉得我不应该不得不切换到一个专门用于编程AI的界面。

I felt like I shouldn't have to navigate to a place where I work with like my coding AI.

然后我还要去另一个地方使用我的销售AI。

And then I have this like different place for my like sales AI.

我还得特意提醒自己：现在我在和销售AI对话，然后开始操作。

And I have to like be like, hey, I'm now talking to sales thing and like do that.

它就该是一个我直接和它说话，它就会帮我解决问题的东西。

It's just like, I just gonna talk to a thing and it's just gonna help.

所以我认为，我们未来会拥有聊天或语音形式的对话式界面，它将成为你与之交谈任何事物的核心，你可以将它添加到任何群聊或其他场景中。

So I think what we're gonna have is that we'll have chat or voice, basically conversational interface will be sort of the the pillar of everything that you can talk to about anything, and that you can add into any group chat or whatever.

它能够自行学习如何帮助你。

So it can like discover how to help you.

但如果你是一个高级用户，对某件事非常擅长，你可能不希望因为必须与另一个人交谈而被中间环节打断。

But then if you're like a power user and you're very good at a specific thing, you probably don't want to be disintermediated by having to talk to another person.

这就像是你有一个执行助理，但你只能通过与他交谈来工作。

It'd be like if you had an executive assistant, but you can only work by talking to them.

这简直太烦人了。

That's like super annoying.

对吧？

Right?

所以在某个时候，你希望直接查看笔记，自己浏览并编辑它们。

So So at some point, you wanna you wanna get to the show notes and, like, look at them yourself and, like, edit them yourself.

对吧？

Right?

你想自己编辑内容。

You wanna edit the thing yourself.

所以我认为我们会把聊天功能与专门为个人需求定制的功能性图形界面结合起来。

So I think we'll pair chat with, like, functional, like, graphical interfaces that are bespoke to, like, what someone needs.

就我而言，我可能会通过聊天来完成我的播客准备工作。

So, like, in my case, I will probably chat to, like, do my, you know, podcast prep.

但当我需要查看产品和代码时，我更希望进入像Codex这样的应用，深入操作。

But when it comes to like, actually looking at product and code, I probably want like the Codex app that I can go into and get deep in.

对吧？

Right?

而如果是一个市场人员，他们可能会通过聊天来询问产品相关的问题。

Whereas maybe if we're talking to a marketer, maybe that marketer will like chats to ask questions about the product.

他们不会为了问产品问题就去下载Codex应用。

They're not going to download the Codex app just to ask questions about the product.

但他们可能会使用一个高度定制的图形界面，比如广告数据分析工具，来深入使用。

But maybe they'll have like a super custom GUI for like ad analytics or something that they go into.

我完全理解。

Totally get that.

而且这在某种程度上替我假定了一段旅程中某个时刻的消费者互动。

And it it kind of wrongly assumes on my behalf a consumer interaction at some point in that journey.

我想问你，你怎么看待代理与代理之间的体验，以及如何为代理设计体验？

And I wanna ask you, how do you think about agent to agent experiences and designing experiences for agents?

我们之前讨论过，比如如何帮助大型企业。

We spoke about, for example, going to large enterprises and how you can be helpful.

我只是选了个最无聊的例子：费用审批。

I'm just choosing the most boring thing ever, expense approval.

你可以有一个代理代表我提交我去旧金山旅行的费用，而另一边的代理则代表OpenAI的合规部门进行审批。

You could have agent submission of expenses on my behalf for my trip to San Francisco, and then the agent on the flip side doing approvals for that from OpenAI's compliance department.

你怎么看待这种模式的转变？

How do you think about that and that paradigm shift?

这很有趣。

That's interesting.

说实话，我不确定这会是什么样子。

To be honest, I'm not sure what that's gonna look like.

我最快的回答是，我们在构建编码器时发现，最适合编码器工作的界面，通常也最适合人类使用。

My quickest answer to this is that we've noticed as we build codecs that the best interfaces for codecs to do work are also tend to be the best interfaces for humans.

所以当人们问，如何让我的代码库对代理更高效时？

So when people ask, oh, how can I make my code base, like, more efficient for the agent to work with?

答案通常是：那你自己看过吗？

The answer is often like, well, have you looked at it yourself?

它对人类来说容易使用吗？

And is it is it easy for a human to work with?

一个具体的例子是，在代码库中运行测试。

So like a very specific example would be, like, running tests in a code base.

如果只是简单地设置大多数测试运行器，它们会输出所有测试的全部结果。

Naively, if you just like set up most test runners, they just like in it all the outputs of all the tests.

作为人类，这真的很烦人，因为你必须去找到那个失败的测试，还得阅读成千上万行内容。

And so like as a human, it's really annoying because you have to go in and like find the one that failed, and it's like you've got to read hundreds of thousands of lines.

结果发现，这对AI来说也很糟糕。

Turns out that's terrible for AI as well.

但如果你只输出失败的测试，对人类更好，对智能体也更好。

But if you filter it down to just only emit the failed test, better for humans, also better for agents.

因此，智能体之间的交互点很可能与有人参与时非常相似。

So probably the agent to agent interaction points will be very similar to like if there was a human in the loop.

这很好，因为它意味着你可以逐个替换各个系统。

And that's nice because it means you can kind of atomically replace individual systems.

我提到了我们在LinkedIn上发布的节目，以及来自另一家公司的优秀投资人。

I mentioned our show on LinkedIn and a wonderful investor from a different company.

那是哈利·波特里的伏地魔，你知道的？

Is that Harry Potter, you know, Voldemort?

就像那个不能说名字的人。

And it's like, you know, he who shall not be named.

我不希望萨姆杀了我。

I don't want Sam to kill me.

但来自另一家公司，我曾经

But from another company, I was

比如，你问他，你怎么看待

like, you ask him, how do you think about

编码数据护城河？

a coding data moat?

Anthropic现在拥有所有数据了吗？

And does Anthropic have all the data now?

我认为，根据我们所看到的，当然，这方面我得听从我的研究团队的意见，但我觉得我们拥有足够多的数据来构建非常优秀的编码模型。

I think that from what we've seen, and, you know, and I'll I would defer to my research team on this, but I feel like we we feel like we have plenty enough data to build really good coding models.

实际上，我认为现在获取数据更有意思的地方在于，当我们进入知识型工作任务时，这类数据在互联网上大多数地方其实并不可得。

I actually think the the place that's more interesting for getting data now is, like, as we get into, knowledge work tasks, that's kind of data that's, like, not really, like, available most places on the Internet.

因此，你会开始产生一些非常有趣的头脑风暴，思考如何帮助模型擅长这些任务。

And so you start to have, like, really interesting brainstorms for, like, how to help a model be good at it.

比如，你可能需要付费让人模拟执行任务，以便你能够学习这些轨迹供模型使用。

Like, maybe you have to, like, pay people to, like, simulate doing tasks so that you can, like, learn these trajectories for the model.

也许你应该收购那些已经停止运营但拥有大量数据的初创公司，比如Slack之类的。

Maybe you should acquire startups, you know, that are no longer in business but have have a lot of, like, data, like, say, Slack or something.

是的。

Yeah.

我认为这类知识工作任务的分布比编程要难得多。

I think that that kind of knowledge work task distribution is, like, much harder than coding.

你提到的那种不存在的数据，这一点非常有趣。

That's so interesting you said there about kind of the data that doesn't exist,

换句话说。

so to speak.

你觉得

How do you think

你与数据提供者——你的Macaws、你的Turing、你的Invisibles——的互动如何？

about your interactions with the data providers, your Macaws, your Turing's, your Invisibles, your of the world?

比如，你会在那里投入10倍的资源吗？

Like, will you spend 10x there?

还是你会认为，我们在数据上的花费太多了，应该自己来做数据采集？

Or will you go, we are spending too much on data, we should do it ourselves and do data acquisition?

是的，我的意思是，我们思考这些问题的方式就是，如何尽可能快地推进？

Yeah, I mean, I think the way that we think about these things is just like, how do we move as quickly as possible?

因此，自己搭建这些系统在时间上非常昂贵，而我们是一个小团队。

And so becoming able to set these things up in house is like very expensive in time, and we're a small team.

所以到目前为止，我观察到的是，如果我们需要大规模开展数据项目，通常会寻求这些公司的帮助。

So what I have observed so far is that if we need to run a data campaign at scale, we're usually gonna enlist help from one of these companies.

在Codex的消费者端，我们已经讨论过如何进入企业市场，以及如何提升开发者体验和开发者关系。

On the consumer side for Codex, we've spoken about like enterprises and going into them how to engage in terms of developer experience, developer relations.

在未来一两年内，你们会在低端消费者市场与Lovable和Rapid竞争吗？

Do you compete with a Lovable and a Rapid on a like low end consumer basis in a year or two's time?

这是一个你们会认为，像柯达这样的产品，并不适合每个人来创建个人简介，或小企业搭建自己的网站的业务吗？

Is that a business where you're like, you know what, Kodak is not for every person to create an about me or a small business to create their own site.

你们如何看待消费者端的这种定位？

How do you think about consumer in that way?

是的。

Yeah.

我会说，目前我们并没有感受到直接的竞争。

I would say that right now, it doesn't feel like we're competing super directly.

但我不知道你有没有看过我们的超级碗广告，它的标语是‘你可以轻松构建东西’。

But I don't know if you saw our Super Bowl ad, the tagline of which is this you can just build things.

通过这款应用，我们注意到许多技术能力较弱的人已经开始构建东西了。

With the app, we noticed that like many people who are less technical are starting to build things.

因此，他们所构建的东西大多非常基础，像是‘Hello World’级别的。

And so the kinds of things they're building are much more hello worldy.

所以我认为，我们会看到一些使用场景上的重叠，比如人们因为ChatHPT内置了这些工具而直接使用编解码器。

And so I think that we will see some overlap in use cases where you have people just pulling up codecs because they have it as part of their ChatHPT.

实际上，上周有一个重大公告：我们现在甚至向免费版ChatHPT用户或Go版ChatHPT用户提供一些编解码器。

Actually, like a big announcement, last week was that we're now offering some codecs to people even on free ChatHPT plans, or on the Go ChatHPT plan.

这在让更多人获得访问权限方面意义重大。

So this is this is massive just in terms of like bringing availability to everyone.

所以我认为，肯定会有一些使用免费 ChatHPT 计划的人进来，构建一些原本他们可能会去使用专业工具完成的简单东西。

And so I think we're definitely going to see people with like a free ChatHPT plan coming in and just like building simple things where they otherwise might have gone to a specialized tool.

你最想做但因为某种原因无法改变的是什么？

What would you most like to do differently, but for whatever reason you can't?

我觉得最近几周对我们来说非常好。

I feel like it's been a very good few weeks for us.

所以我们非常兴奋，对正在发生的一切都充满热情。

So we're very I'm pretty jazzed by everything that's happening.

我感受到的最多的就是这种情绪。

Feeling that I have the most.

是的。

Yeah.

这真的很有趣。

That's really interesting.

你说最近几周

You said it's been a

对我们来说，这是非常棒的几周，我也感受到了。

very good few weeks for us, and I feel that.

团队是否感受到正负两方面势头的变化？

Does the team feel the changing winds of momentum both in positive and negative cycles?

当然。

Absolutely.

我们对此非常敏感。

We are very attuned to it.

对吧？

Right?

比如，看看Codex的历史，我们去年推出的第一个东西，是一个让人无比兴奋的绝佳创意。

Like, you look at the the history of Codex, the first thing we launched last year was, like, this amazing idea that people were super excited about.

就像是，我们给智能体在云端配备一台独立的电脑，你可以拥有任意数量的智能体并行为你完成任务。

It's like, hey, we're going give the agent its own computer in the cloud, you can have as many of them as you want work for you in parallel on tasks.

超级棒的点子。

Super great idea.

说实话，它没有我们后来发布的产品那么好。

To be honest, it didn't work as well as what we shipped later.

它并不是最好的。

It was not the best.

从八月开始，随着GPT-5的推出，我们大力推动交互式编程，这也是市场上大多数竞争的焦点。

And then since August with GPT-five, we started pushing really hard on interactive coding, which is where most of the competition in the market is.

我们简直势如破竹。

You know, went on an absolute tear.

我觉得我们当时的公开数据是，自八月以来，我们的增长达到了20倍。

I feel like the public metric we had was like since August, we grew by like 20 x.

甚至在年底，我们从十二月到现在又翻了一倍。

And then like even like late in the year, we like doubled from December to now.

那个确切的数字我记不清了。

I forget the exact number there.

但那确实是与对手激烈竞争、不分上下。

But, like, that was competing neck and neck.

但上周我们感受到的转变是，我们认为自己拥有最智能的模型，这一点通过五三编码器得到了巩固。

But the the shift that we feel last week is, you know, we we felt like we had the most intelligent model that was cemented with five three codecs.

我们收到了一些反馈，说我们的模型运行较慢，可能不够有趣，而且在工作时与用户沟通的能力也较差。

We had feedback around our model being slower and, like, maybe less fun to work with and, like, being less good at communicating with you while it was working.

我们针对这些反馈进行了改进，甚至相比那些比我们早二十分钟发布竞品模型，情况也明显更好，或许可以说它们有点逊色。

We addressed that feedback, and that's true even compared to, like, the the other competitor model that launched, like, twenty minutes before us and was like, maybe this is spicy.

它只领先了二十分钟。

It was, soda for twenty minutes.

Soda 意思是业界领先。

Soda means state of the art.

我们一直收到很多关于 Codex 用户体验质量的反馈。

And then we'd always been getting a lot of feedback on, like, the quality of the user experience in Codex.

我们最受欢迎的界面是 IDE 插件，而命令行界面（CLI）则显得不够完善。

Our most popular surface was the IDE extension, and our CLI, which is the command line interface, was less polished.

但通过这款应用，市场反馈强烈表明，这是一款非常高质量的体验。

But with the app, the feedback has been like resounding from the market that this is like a really high quality experience.

它简单得有点出人意料，人们都非常喜欢使用，即使是我们最大的用户也已全部转化。

It's like simple, like unintuitively simple, and people are just loving using even our biggest credits are converted.

所以是的，然后我们做了超级碗广告，接着就免费了。

So yeah, and then we and then we had the Super Bowl ad, and then we went to free.

回到你之前问的问题，我最想做出什么不同的改变？

And going back to your question of like, what do I most want to do differently?

我有两点想跟你分享。

I have two things for you.

第一点是，我其实想重新回归云服务。

The first is I actually want to get back to cloud.

当我们去年将战略从专注于云代理转向交互式开发时，当时的思路非常简单。

When we pivoted our strategy from like focusing on the cloud agent last year to working interactively, the thinking was very simple.

这其实有点像我之前跟你提到的FDEs的情况。

It was just and it's kind of like what I was telling you about FDEs actually.

如果你在终端用户还没熟练掌握工具、无法轻松使用之前，就过早地追求工作流自动化，就会出现脱节，最终只能是一个不切实际的空想，除了最资深的用户外，其他人根本用不起来。

If you go too far ahead to workflow automation before your end user is fluent with the tooling and can get it to work simply, then there's like this disconnect and you just have this pipe dream idea that's not like effective except for the most power users.

但一旦你有了一个基础，人们每天都在使用你的工具并进行配置，每次使用它时它都会变得更好，那么让其在云端独立运行就只是一个很小的跃升。

But once you have this base where people are using your tool every day and they're configuring it and every time they use it, it gets better, then, like, the step up to, like, letting it run independently in the cloud is a much smaller step up.

所以我认为现在是我们重新回归云端产品开发，并使其与本地产品高度集成的时候了。

So I think it's time for us to, like, get back to, like, building out the cloud product and making it super tightly integrated with the local product.

它目前已经有一定的集成度了。

It already is somewhat integrated.

我想做出的另一个改变是，更多地关注瓶颈问题。

And the other thing I want to do differently is, start thinking more about the bottlenecks.

比如代码生成，写代码现在基本上已经变得微不足道了。

Like code gen, writing code has become, like, basically trivial now.

但困难的部分正如你提到的代码审查，对吧？

But the hard part is like what you were talking about with like code review, right?

我们如何知道代码质量是好的？

Like, how do we know the code quality is good?

我们如何确定自己在做正确的事情？

How do we know we're doing the right things?

这些瓶颈仍然被低估，投入也不够。

And those bottlenecks, I think are under underappreciated still and under invested in.

所以，我认为我们要朝着这样一个方向努力：拥有一个不受瓶颈限制、值得信赖的代理，它可以独立负责整个微系统或内部工具，并完成完整的迭代循环，包括来自用户的反馈，而无需人工审查。

So like, I think we want to get to a world where you can have an agent that is unbottlenecked that you trust to like own an entire microsystem or internal tool or whatever, and can do the full iterative loop, including feedback from users without having to go through human review.

这既是一个从智能角度难以解决的问题，也是一个从安全性和控制角度难以解决的问题。

And that is a really hard problem to solve both from an intelligence perspective, but also from like a safety perspective and a controls perspective.

我们应该在基准测试和评估上赋予多大权重？

How much weight should we place on benchmarks and evals?

这可能是个让你觉得烦的答案。

Probably, this is an annoying answer for you.

就是一些吧。

It's like some, right?

在我看来，它们确实能很好地衡量智能水平。

Like they do tell you in my mind, they give you a good measure of intelligence.

因此，你可以对这些基准在智能方面的表现给予重视，尤其是在评估指标尚未饱和之前，当你看到这些基准取得实质性进展时，会非常有帮助。

And so you can put weight on those for intelligence And especially before evals are saturated, I think when you see meaningful progress in those benchmarks, it's like very, very helpful.

然后我认为你必须将这一点与使用模型的实际感受结合起来。

And then I think you have to pair that though with like what it feels like to use the model.

这是一种感觉上的东西。

And that's that's a vibes thing.

每当我与任何人交流，哪怕是内部人员，或是我们模型的客户时，我总是惊讶于人们对模型使用体验的评估竟然如此依赖感觉。

Like whenever I talk to any, even internally or even talking to, like, customers of our models, I'm always surprised by how vibes based the evaluation of how it feels to work with the model is.

生活真是靠感觉啊。

How vibes based life is.

人们更愿意和自己喜欢的人共事，这是我给孩子们的建议。

People wanna work with people they like is the lesson that I give to kids.

没错。

Exactly.

人们更愿意和自己喜欢的妈妈们共事。

People want to work with the moms they like.

从市场构成的角度来看，作为投资者，我必须思考如何预测这个市场的最终状态。

In terms of like market composition, as an investor, I have to think through how do I think about the eventual state of this given market kind of a terminal state.

你如何看待这一点？

How do you think about that?

是像Uber和Lyft那样，大部分市场都会集中在Codex或ClawCode上吗？

Is it like Uber and Lyft and like the majority of the market will be on Codex or ClawCode?

还是像AWS、Azure、Google Cloud那样，形成三足鼎立的局面？

Or is it like AWS, Azure, Google Cloud and a 30 three-thirty three-thirty three?

好吧，我认为从长远来看，最终能捕获大量价值的提供商可能会更少。

Okay, so I think this might end up with fewer providers that are capturing a lot of value in the long run.

原因如下。

And here's why.

这可能有点尖锐，但我认为我们现在正处于一个临时阶段，即代理在编程方面表现得非常出色。

And maybe this is a bit spicy, but I think that we are kind of in this temporary phase where we have agents that are really good at coding.

对吧？

Right?

如果回顾去年，也许更多人认为我们会拥有在其他领域也表现优异的代理。

And if you look back last year, like maybe more people thought we would have agents that are good at other domains too.

但去年并没有发生这种情况。

But that didn't happen last year.

所以目前来看，整个行业只有编程代理达到了产品市场契合度，对吧？

So we only have PMF for coding agents, like in the industry overall, I would say, right?

然后还有一些非常狭窄的其他应用场景，比如客户服务等。

And then there's some like very narrow, narrow other use cases, like customer support, etc.

但我认为这可能是暂时的。

But I think that's probably temporary.

随着时间推移，我们将拥有能够为你做任何事情的代理。

And then over time, we're going to end up with agents that kind of can do anything for you.

这就像我之前说的，会有一个超级助手，你可以跟它聊任何话题。

This kind of what I was saying earlier, like there's just like a super assistant, you talk to it about anything.

然后如果你恰好深入某个特定功能，也可以使用专门的界面查看。

And then there is like specific UI that you can go look at if you happen to be deep in a specific function.

在这种情况下，我不认为一家公司需要配备12个不同的代理，让员工去摸索该找哪个，因为那样他们就无法达到熟练程度。

So in that world, I don't think you want like 12 agents at the company and you have to like go your employees have to go figure out the right ones to talk to, because then they won't achieve fluency.

如果他们不想达到熟练程度，就不会愿意将自动化融入自己的工作中。

If they don't want to achieve fluency, then they will also won't like pull automation into their roles.

但如果你有一个可以和它聊任何事情的工具，对吧？

But if you have this one thing that you can talk to about anything, right?

所以在入职时，只需去和这个工具聊聊你需要的任何事，人们就会形成习惯去使用它，它将成为工作的中心，人们也会主动引入自动化。

So you're onboarding, it's just like go talk to this thing about anything you need, then people will develop muscle memory to go to it, it'll become the center of gravity of work, and people will pull an automation.

所以我认为，这种未来更有意义。

So I think that that future makes much more sense.

而且我觉得，像ChatGPT的构建者们，其实非常适合实现这一点。

And I think like, as the people building ChatchPT were like really well set up to deliver that.

这可能有点牵强，但一个类比是，我以前在Dropbox工作过。

This this is kind of a stretch, but an analogy here is I used to work at Dropbox.

在Slack还没兴起的那段时间里。

And for a while, this is before Slack was big.

有一段时间，我们一直在想，人们应该是在Dropbox的文档里评论，还是应该去Slack里讨论这些文档。

And for a while, we thought we wondered if people should like go comment on like documents in Dropbox or or if they should like go talk about the documents in Slack.

很明显，人们更倾向于在Dropbox中视频的正确时间戳下留言，或者直接在文档上留言。

And it was like obvious that it was like more optimal for people to like put comments on the right timestamp in the video in Dropbox or like comment on the document in Dropbox.

对吧？

Right?

所以这样更高效。

So it was more optimal.

然而，我们看到Slack已经成为人们彼此交流的中心枢纽。

However, what we saw is that Slack is just such a center of gravity of people just like talking to each other.

根本没人愿意在文档上留言。

Like nobody wants to comment on the document.

我只想在Slack上给你发消息。

I just want to Slack you.

因此我们发现，即使效率更低，人们也强烈倾向于在Slack上进行交流。

And so we saw that like there was this really big pull towards things happening in Slack even if it was less efficient.

我认为工作中也会出现类似的情况：如果有一个能应对几乎所有需求的单一智能体，就会形成巨大的吸引力，每个人都会讨论他们如何用这个智能体处理各种事情。

And I think we're going to see something similar at work where if there is a single agent you can use for nearly anything, it there will just be this giant poll, and everyone will talk about how they use that one agent for things.

团队之间会分享最佳实践。

Teams will share best practices with each other.

会围绕如何最佳使用这个工具举办黑客马拉松。

There'll be hackathons around how to use that best thing.

是的。

Yeah.

最终你只会剩下寥寥几个这样的工具。

And you'll end up with just a handful of these.

你提到过，除了编程之外，代理在其他领域的使用并没有真正扩散，而客户服务就是一个例子。

You said about kind of agents not really proliferating in terms of usage other than coding, and actually maybe this being the time and and, customer support is one of the examples.

我的问题是，我今天是一名投资者。

My question to you is, I'm an investor today.

我在寻找那些能够随着时间积累价值、为客户提供卓越产品的企业。

I'm looking for companies which would accrue value over time and provide incredible products to customers.

有一种观点认为，如今大型SaaS公司的收入持久性为零，SaaS已经死亡，因为像你、Anthropic这样的模型提供商将前来分走我们的蛋糕。

There is a belief that the durability of revenue of large SaaS companies today is zero and that SaaS is dead because the model providers, you, Anthropic, others are going to come for our lunch, so to speak.

你建议我怎么做？

What would you advise me?

像这些工具都是为人类设计的，否则意义何在？

Like things are built for humans, like otherwise, what's the point?

即使是SaaS工具也是为人类设计的。

Even SaaS tools are built for humans.

所以对我来说，我想问的是，这家SaaS公司是否与终端的用户建立了关系？

So for me, I think my question is like, does this SaaS company own a relationship with a human on the other end of things?

如果建立了关系，那我怀疑它不会消失？

And if it does, then I suspect it's not going away?

还是说这家SaaS公司掌握着某种至关重要的核心数据系统？

Or does the SaaS company own some like really important system of record?

它大概也不会消失。

It's probably not going away.

也许这两点——与人的互动和核心数据系统——实际上比以往任何时候都更重要。

Maybe both of those two things, the interaction with the human and the system of record are like more important than ever actually.

另一方面，这家SaaS公司是否只是一种粘合层，却并不拥有上述任何一项？

On the other hand, is the SaaS company like a kind of a glue layer, but it doesn't own either of those two things?

我不是这方面的专家，但我对这类公司更感到担忧。

I'm not the expert here, but I'm more nervous about that kind of company.

如果我们持这种观点，Salesforce和ServiceNow的股价已经下跌了20%、30%、40%。

If we take that stance, Salesforce and ServiceNow, they're down 20%, 30%, 40%.

我认为这种下跌被严重夸大了。

I think it's massively exaggerated.

我认为确实有一些公司理应如此。

I think there are some companies that legitimately should be.

恕我直言，我认为Dropbox正处于非常艰难的境地。

Respectfully, I think Dropbox is in a very difficult position.

但像monday.com这样的公司，对于绝大多数使用它的中小企业和消费者而言——这实际上是它们的主要市场——难道用户不能自己用代码写一个待办事项清单吗？

And I think your monday.coms of the world though, for the majority of SMBs and consumers who use it, which is the large majority of their market, actually, could they vibe code a to do list?

当然可以。

Yes.

这样做会更经济高效吗？

Would it be cost efficient to do so?

其实不会，一旦你开始定制并完善它，成本就上去了。

Not really actually by the time you customize it and perfect it.

说实话，这个待办事项列表在功能上通常都很平淡，无非就是添加任务、完成任务、显示分配给新成员的历史任务。

And to be honest, this to do list is generally pretty bland in terms of what you need to do, add task, complete task, show historical tasks assigned to new members.

这并不难。

It's not very difficult.

所以实际上，我觉得你就直接保留它好了。

And so actually, I think you just keep it.

所以我觉得这被严重夸大了。

And so I think it's massively overblown.

我认为这是市场典型的本能反应。

And I think that's the classic knee jerk reaction from markets.

我觉得你最终还是会转向客户支持，而我不希望自己属于这一类。

I do think you're going to come for customer support and I wouldn't want to be in that category.

我觉得这可能会改变你投资的创业者类型。

I think this maybe changes what kind of founder you invest in.

对吧？

Right?

比如，曾经有一段短暂的时期，作为产品构建者，我个人很喜欢那种专注于能做出好产品的创业者。

Like, think there was this maybe temporary phase where that I liked personally as a product builder, there was this phase where you would invest in like the the person who can just like build good product.

当时你可以忽略他们是否对客户、市场推广或分销有清晰的见解，因为做出好产品实在太难了。

And you could kind of ignore if they had a good thesis around a customer or go to market or distribution or anything like that, because it was so hard to build good product.

对吧？

Right?

我认为那只是一个异常现象。

And I think that was a that was an anomaly.

如果我们看看现在的情况，也许这种类型的创业者不再是你应该投资的对象了，因为现在做出好产品相对容易了，你需要重新回归到投资那些深入思考过分销渠道、对特定客户群体有深刻领域专长的创业者。

If we look at where we are now, like, maybe that kind of founder is not the founder you should invest in because it's, like, kind of relatively easier to build good product, and you need to go back to, like, investing in the founder who's, like, thought through distribution, who has a good good domain expertise of what to build for a specific customer, etcetera.

所以，再次假设你是我团队中的一名投资人，你会如何思考我们值得关注的领域，投资那些能够积累价值、不会被模型提供商威胁的公司？

So, again, if you were on my team as an investor, how would you think about interesting areas for us to invest in companies that will accrue value and not be threatened by model providers?

因为again，你进入医疗领域，进入代码领域，显然Codex非常明确，你进入客户服务领域，那你不去哪里呢？

Because again, like you're going to health, you go into code, obviously, Codex is very clear, you go into customer support, Where are you not going?

那为什么Claude Code不去呢？

And why is Claude Code not going?

我真想直接说，我不知道。

I'm tempted to just say, like, I don't know.

我觉得现在当投资人真不容易。

I couldn't I I think it's a hard time to be an investor.

市场太动态了。

It's a the market is so dynamic.

很难说清楚。

It's hard to say.

如今投资真的非常困难。

It's a really tough time to be investing today.

我的回答其实是两方面的：第一，我寻找具有物理基础设施的领域。

My my answer is kind of twofold actually, which is like, number one, I look for things with physical infrastructure.

我不认为你会进入能源供应领域。

I don't think you're going into energy supply.

第二点是金融科技和银行整合，以及复杂的金融产品。

And then two is like the fintech and banking integrations, gnarly financial products.

我不认为OpenAI会去东南亚的500家银行建立合作关系。

I don't think OpenAI is going to go into building 500 relationships with banks in Southeast Asia.

我倾向于同意。

I tend to agree.

我再说一遍，这又回到一个问题：你是否要进入一个极其复杂、客户关系和市场知识至关重要的市场？

I I I again, it comes back to, are you going into, like, a gnarly, complicated market where customer relationships and, like, knowledge of the market are everything?

这看起来仍然很不错。

That still seems great.

人才争夺战有多激烈？

How bad is the war for talent?

从英国来看，我们观察旧金山，我会告诉企业，最好在欧洲发展，因为根本无法吸引和留住人才。

From The UK, we look at SF, and I say to companies it's better to build in Europe because it's impossible to acquire talent and it's impossible to retain it.

我错了吗？

Am I wrong?

我认为现在的人才争夺战极其激烈。

I think that the war for talent is incredibly fierce right now.

你知道，显然在OpenAI，我们拥有非常强大的品牌，因此能够吸引大量人才。

You know, obviously, at OpenAI, we have an incredibly strong brand, and so we're able to attract a lot of talent.

但即便如此，我们仍投入大量精力去争取那些我们非常看好的候选人。

But even so, we put a ton of effort into, like, closing candidates that we're really excited about.

甚至可以说，我们也不觉得可以免费随意招到任何想要的人。

Even, like, even we feel that it's not like you don't just get whoever you want for free.

我可以问一下，你们提供股票的入门价格，对顶尖人才来说仍然有吸引力吗？

Can I ask, at the entry price that you get a stock at, is it still attractive for the best talent?

还没有人告诉我相反的情况。

I haven't had anyone tell me anything to the contrary.

你在多大程度上会考虑寻找完美匹配的人，而不是找一个足够好的人？

To what extent do you think about, like, finding the perfect fit versus finding someone who's good enough?

所以，你知道吗，我之前开过一个玩笑，说产品经理其实是可有可无的。

So, you know, earlier I made my joke about, like, PMs kinda being optional.

是的。

Yeah.

但我认为这其实并不对。

I think that's not actually true.

你仍然需要产品方面的人才。

You still need product people.

但我确实认为他们必须是完美匹配的人选。

But I do think that they have to be the perfect fit.

如果你找的人不是完美匹配，反而可能弊大于利。

And if you have someone who's like not the perfect fit, might just do more harm than good.

这意味着我们在选拔上比我在其他职位时要严格得多。

It's kind of means that like we're way more selective than I might have been in other roles.

我是一名计算机科学专业的学生。

I'm a CS student.

好吧？

Okay?

我在斯坦福。

And I'm at Stanford.

我是帝国理工的。

I'm an Imperial.

我在剑桥。

I'm at Cambridge.

我在ETH任何地方。

I'm wherever ETH.

顶尖学府。

Great institution.

基于你现在的所有认知，你对我有什么建议，能帮助我规划未来五年的职业发展？

What would you advise me knowing all that you know now that would help me navigate the next five years of my career?

我想成为一名工程师，在未来一年进入职场，并为AI生态系统创造价值。

I want to be valuable to the AI ecosystem environment as an engineer entering the workforce in the next year.

基本上，现在是工程师最好的时代，因为你拥有令人难以置信的工具，可以完成海量工作。

Basically, there's actually never been a better time to be an engineer because you have incredible tooling available to you to get incredible amount done.

而且，你快速上手一个可能被雇佣进入的复杂代码库的能力从未如此之快，因为你随时可以向AI询问大量关于代码库的问题，还能让它帮你规划那些原本可能需要你花几天时间研究的改动。

And your ability to like ramp into like a complex code base that you might be hired into has never been faster because you can go ask AI like a ton of questions about the code base, and you can ask it to plan out changes that would otherwise take you like days to research maybe.

我认为首先，你应该保持非常乐观的态度。

I think first off, I would say like you should be like very optimistic.

但当然，你也要考虑自己入职后希望提升哪些能力。

But then of course, like about you want your abilities once you're at the job.

那么接下来的问题就是：你该如何找到这份工作？

Then now the question is how do you get the job?

因为现在构建东西从未如此简单。

Because it's never been like easier to build things.

真正变得稀缺的是主动性、品味和质量。

The thing that becomes scarcer is like agency, taste, and like quality.

我建议你直接动手去构建一些东西，展示你的主动性和对作品的品味，打造高质量的作品，然后分享出去。

I would urge you to like, just build things and demonstrate your agency and your taste around what you build and like build things that are of high quality and then share those things.

你知道吗，我们收到了很多来自申请职位的人的主动联系，他们有的通过招聘页面申请，有的则通过社交媒体联系。

You know, we get a lot of inbound for from folks both applying for jobs through the careers page or also on social.

这只是我个人的看法，但当有人给我写信，分享一些有趣的想法，并附上一个有趣项目的链接时，这比一份普通的简历更能引起我的注意。

And this is just me, but when someone writes to me with like some interesting thoughts and, like, a link to an interesting project, that gets my attention much more than, a normal resume does.

在进入快速问答环节之前，最后几个问题。

Final questions before we do a quick fire.

你之前提到了Dropbox。

You mentioned Dropbox earlier.

Dropbox的校友群体非常出色。

The alumni from Dropbox is incredible.

我的意思是，看到从Dropbox走出来的这么多人才，真的令人惊叹。

I mean, really, like, amazing to see the talent that's come out of Dropbox.

你在Dropbox学到的最重要的一个经验是什么？这个经验如何影响了你现在在OpenAI的思维方式？

What was your single biggest lesson from Dropbox that has shaped some of your thinking now with OpenAI?

哦，这个问题我都不用想。

Oh, I I don't need to think about that one.

这正是我之前跟你说的那件事。

That that's kind of the thing I was telling you about earlier.

对吧？

Right?

当你为人们、为终端用户构建工具时，你必须把这种工具视为一种参与系统，对吧？

Like, when you're building tooling for people, like, for end users, you have to think about, like, that tooling as a system of engagement, right?

如果人们不想用你的工具，如果它不能自然地成为完成某件事最简单的方式，那么人们就不会使用它。

If people don't want to use your tool, if it doesn't like naturally feel like the easiest way to get something done, then people just won't use it.

我从观察Slack如何迅速火爆中学到了这一点。

Again, I learned that from watching how Slack just absolutely took off.

所以现在我们在构建这些智能体时，经常会想到这一点：如果我们只是把我们的智能体做成工作流自动化，那每次想让它启动都会像拔牙一样困难，对吧？

And so I think about that a lot now, when we're building these agents, I'm like, if we build our agent surely is like, workflow automation, then it's always going to be like pulling teeth to get that thing started, right?

你得雇咨询公司比如埃森哲之类的人来介入，他们还得部署全职人员，这会非常困难。

You're going to need to hire Accenture or someone to come in, they're going to deploy FTEs, it's going to be tough.

但如果你能构建一个让人爱不释手的系统，即使他们只用它来完成部分任务，随着时间推移，他们会越来越熟练，也会逐渐将它与你希望集成的工具连接起来，然后你就能逐步引入自动化。

But if you can build a system that, like, people just love using, even if they only use it for partial tasks, over time, they'll get better and better at using it, and then that you'll get connected to the tools you want over time, and then you can start laddering in automation.

显然，这些并不是互斥的。

Obviously, these aren't mutually exclusive.

你如今该如何重振Dropbox的增长？

How on earth do you reinvigorate growth at Dropbox today?

至少在我还在Dropbox的时候，我们最擅长的是桌面软件。

At least from when I was at Dropbox, the thing we were uniquely good at was desktop software.

而桌面软件，说来有趣，它从未真正退出过。

And desktop software, it's funny, it was never not back.

但不管怎样，它现在又火了。

But anyways, it's so back.

基本上，因为如果你在解决生产力和知识工作的问题，确实，到处都有你需要连接的数据系统。

Basically, because if you're solving for productivity and knowledge work, yes, there are systems of record everywhere that you need to connect with.

但归根结底，所有工作最终都发生在用户的电脑上，要么在浏览器里，要么就是本地的应用程序。

But everything at the end of the day happens on the user's computer, either in their browser or just like locally and apps on their computer.

我认为，我们看到AI代理在工作中带来生产力提升的最快方式，首先是让它们直接在用户的电脑上与他们已有的工具协同工作，而无需部署FDE来设置任何东西。

I do think that the fastest way we're going to see productivity gains from agents at work is going to be at first meeting users on their computer, working with the stuff that they have available to them, you know, without having deployed FDEs to set anything up.

随着时间推移，你会逐步接入这些不同的系统。

And then over time, you'll connect in these various systems.

所以如果我是Dropbox，我会思考如何利用我们在构建优秀桌面软件以及计算机上协作层方面的独特专业能力，来赋能生产力代理工具。

And so if I was Dropbox, I'd be thinking about how do we leverage our unique domain expertise in like building really good like desktop software, and the sort of collaborative layer on top of your computer, how do we leverage that to enable productivity agents?

这有点宽泛，但我认为这就是你要切入的角度。

It's a bit broad, but I think that's the angle you go.

不，我非常喜欢这个回答，也非常感谢你的分享。

No, I love it and I really appreciate the response.

在我们进入快速问答之前，最后一个问题是。

Final one before we do a quick fire, I promise.

我成长于一个重视利润率的世界。

I've been brought up in a world where margin matters.

软件的利润率非常可观，这也是软件成为绝佳投资领域的原因。

Software margins are wonderful and it's what makes software a brilliant category to invest in.

我们正看到，在以推理为主的业务中，利润率状况变得非常不同。

We're seeing margin profiles that are very different in inference heavy plays in particular.

我该多大程度上忽略这一点，相信成本会下降，令牌成本会下降，真正重要的是使用量和用户喜爱，利润自然会来，还是说利润其实极其重要？

To what extent should I put that out of mind and appreciate that costs will come down, cost of tokens will come down and actually it's about usage and customer love, margins will come or no, margins are actually freaking important.

保持这个焦点。

Keep that focus.

我认为两者成本都会大幅下降。

I think both costs are gonna come down significantly.

而且我认为，如果今年是代理在工作中广泛部署的一年，那么今年也是它们必须连接到所有这些系统的年份。

And I also think that if this is the year of agents being deployed like broadly at work, then this is also the year where they're going to have to be connected to all these various systems.

我认为这会非常具有粘性。

And I think that's going to be very sticky.

所以我认为今年是一场竞赛。

And so I view this year as a race.

因此，你希望赢得这场竞赛，同时暂时承受一些利润损失也是可以接受的。

And so I think you want to win that race, and you should be okay taking some hit to margin in the meantime.

老兄，快问快答环节。

Dude, quick fire round.

所以我先说一个简短的陈述。

So I say a short statement.

你告诉我你最直接的想法。

You give me your immediate thoughts.

这样可以吗？

Does that sound okay?

是的。

Yeah.

在过去十二个月里，你最改变主意的是什么？

What have you changed your mind on most in the last twelve months?

我加入OpenAI的时候，我认为——这比十二个月前还要早一点。

When I joined OpenAI, I thought that and this is a little longer than twelve months ago.

但我加入OpenAI时，以为我们都会只是和电脑屏幕共享内容。

But when I joined OpenAI, I thought that we would all just be hanging out with our computer screen sharing.

但就在那之后的一年内，我们会拥有一个可以和我们对话的智能代理。

But within a year from there, you know, we'd have this agent that we're just talking to.

那完全错了。

That was completely wrong.

我认为多模态模型的进展速度比我预期的要慢。

I think the rate of like progress in like multimodal models was like slower than I expected.

多模态指的是，比如能够处理视频和音频的模型。

Multimodal means, you know, like models that work with like video and audio.

但事实上，我们发现通过代码与你的电脑交互的智能体才是正确的方向。

So instead, what happened was that we saw that like agents that work with your computer through code are the way.

因此，对我来说，这彻底改变了我对如何将人工智能的好处普及给大众的看法。

And so for me, that's been a complete rethink in terms of like how we bring the benefits of AI to like just people generally.

主要并不是通过视频和音频。

It's not not through video and audio primarily.

你最尊重哪个不太为人所知的竞争对手？为什么？

Which lesser known competitor do you respect most and why?

我第一个想到的是AMP。

The first one that came to mind was AMP.

AMP。

AMP.

我认为他们正在构建中。

I think they're building yeah.

AMP。

AMP.

它来自Sourcegraph的团队。

It's out of out of the folks at Sourcegraph.

他们的产品口碑非常好，真正做到了以小博大。

Their product has a great reputation of just being, like, you know, punching way above its weight.

但我真正钦佩的另一点是，他们推动了围绕智能体的整个标准化进程。

But I think the other thing that I really respect is that they helped initiate this whole like standardization around like agents.

Md 和 dot agents/skills，这正是我前面提到的，让用户更容易管理他们正在尝试的各种智能体。

Md and like dot agents slash skills, which are what I was saying earlier about like, making it so that it's easier for users to manage all these different agents that they're trying.

显然，我们发布了智能体。

Obviously We put out agents.

M d，但他们发布了代理。

M d, but they put out agent.

M d。

M d.

基本上，昆恩通过发布一条推文开启了这一切，内容是：嘿。

And, basically, Quinn started this all by putting out a tweet that said, hey.

如果你们购买了 agents 这个域名。

If you guys buy the domain agents.

M d，我们就按照你们的拼写方式标准化。

M d, we'll standardize to your your spelling.

尽管这件事很小，但它引发了这场我認為對社區來說非常棒的标准化进程。

And as small as that was, that initiated this whole standardization that I think has been awesome in the community.

你认为对Anthropic广告的回应是正确的吗？

Do you think the response to Anthropix ads was the right response?

我的意思是，当时有太多不同的回应了。

I mean, there were so many different responses.

我听到的那个，显然觉得是正确的。

The one that I heard, obviously, think was right.

我听到的说法是，一家公司对未来的看法相当消极，而另一家公司，也就是我们OpenAI，则非常积极，告诉人们可以构建属于自己的梦想。

The one that I heard was, well, one company is being pretty negative about the future and the other company, us, OpenAI, is being really positive and just telling people they can build things into Dream.

我觉得这个回应太棒了。

I I thought that response was brilliant.

自从加入Codex以来，你做出过最难的产品决策是什么？

What's the hardest product decision you've had to make since being at Codex?

我可以告诉你我们曾经做出的最痛苦的产品决策。

Well, I can tell you the most painful product decision we had to make.

有一段时间，Codex Cloud实际上是无限制的，虽然不是免费的——你需要为Traci PT付费，但使用量是无限的。

For a while, Codex Cloud was, like, effectively unlimited, not free, like you needed to pay for Traci PT, but then you had unlimited usage.

我们每天维持这种无限使用状态，都知道将来要收回这个政策会越来越难。

Every day that we left it that way, we knew that would be harder to wind back at being like unlimited.

但我们当时太专注于其他更具产品市场契合度的竞争点，所以就把这个决策推迟了。

But we were just so focused on competing on our other things that had more PMF that we kind of punted that decision out.

当我们把无限使用限制为更合理的额度时，用户们产生了大量反弹。

When we wound back that unlimited use to some like more reasonable limit, there was a lot of blowback from users.

但只有极少数用户认为所有东西都该永远免费。

And it was a very small minority of users who like thought everything should be kind of like sudo free forever.

但这种反弹影响了我们方方面面，因为社交媒体上的讨论并不会区分这些细节。

But that blowback affected us everywhere because like the social chatter doesn't really distinguish between these things.

我认为我从中艰难学到的教训是：你不能让东西无限免费太久。

I think the lesson I learned the hard way there is, like, you can't make things unlimited for too long.

数据网站的定价和祖父条款定价真的非常棘手。

Data site pricing, grandfathering pricing is just it's such a hard thing.

今天我们工程或产品上做的哪些事情，五年后你会回过头来说：天啊，我们居然做过这种事？

What do we do today in engineering or product that in five years time, you'll look back on and go, oh my God, can you believe that we did that?

其中之一就是手动编辑代码。

Well, one is just editing code by hand.

我想另一个可能更尖锐的是：手动管理系统的部署和监控。

I think probably another one, this is maybe spicier, but another one might even be like actually managing the deployment and monitoring of systems by hand.

我觉得大型公司可能需要很长时间才能部署这项技术，但许多初创公司可能会直接基于一个全新的、完全由AI管理的栈来构建。

Like, I basically think that probably big companies will take a long time to like deploy this, but many startups might actually kind of start building on a completely new stack that's like fully AI managed.

需要明确的是，这个栈目前还不存在，但一个完全由AI管理的栈将会被设计成对代理的行为提供强大的确定性约束，并能轻松回滚部署等操作。

To be clear, the stack doesn't exist yet, but a fully managed AI stack where because the basically it's been built to give you really strong deterministic guardrails over what the agent can do and like control of to like rollback deploys and everything like that.

因此，未来你创办一家公司的方式可能是：先获取一个代理，然后直接让它帮你构建产品。

And so we'll get to a world where the way you start a company is you start by getting an agent and just asking it to build things.

然后你会拥有更多的代理。

And then you get more agents than that.

最终，你可能会把联合创始人也加入到你用来与代理协作的服务中。

And then maybe eventually you add your co founders to this service that you use to work with agents.

于是，你最终可能会发现，你的主要沟通工具其实是代理沟通工具，你不再需要亲自处理那些繁琐、痛苦的CI和部署流程，而是让代理去完成这些工作。

And so you end up like, maybe your main communication tool is actually your agent communication tool, and then maybe you're not actually handholding this, like, very point painful CI and deploy process, but you're just, like, having agents do things.

有个奇怪的问题，但我很好奇。

Weird question, but I'm intrigued.

你是提供代理约束机制的那个人吗？

Are you the one providing agent guardrails?

我的意思是，你的代理可以在企业内部任意访问。

And what I mean by that is your agents can go anywhere within an enterprise.

那么，提供这些安全限制的是你，还是有第三方服务商说，‘亚历克斯，你不能进入那个区域，那是人力资源部’？

Are you responsible providing those guardrails or is there a third party matter provider who is saying, hey, Alex, you can't go into that, that's human resources.

或者你不能进入那个区域，那是市场部。

Or you can't go into that, that's marketing.

你怎么看待安全限制的提供方式？

How do you think about guardrail provisioning?

这是代理提供商的责任，还是第三方提供商的责任？

And is that the role of the agent provider or a third party provider?

我认为我们可能会看到两种方式并存。

I think we'll probably see both.

就像我说的，我们正在为代理安全限制投入大量精力。

Like we are putting a lot of effort into agent guardrails.

就像我说的，我们几乎是唯一一家重视为编码代理提供操作系统级沙箱隔离的公司。

Like I said, we have, we're basically the only company that cares about OS level sandboxing for coding agents.

例如，Windows 上目前没有任何这样的方案。

For instance, there's none that exists on Windows.

我们正在开发这个功能。

We're the ones building that.

而且我们是以开源的方式进行的，希望其他人也能使用。

And we're doing it in open source, hopefully other people can use it.

我们对此进行了大量思考。

We think about that a lot.

ChatHPT 支持连接器。

ChatHPT supports connectors.

所以，你可以与你的 Google 文档之类的东西进行交互。

So, you know, you can talk to your, like, Google Docs or something.

我们投入了大量精力来设置代理在操作你的 Google 文档时的权限限制。

And we put a lot of effort into guardrails around what the agent can do with your Google Docs.

这只是两个例子。

Those are just two examples.