应用直观：十亿台智能机器 - [商业拆解，第248集]

本期节目由Portrait赞助播出。

This episode is brought to you by Portrait.

Portrait由前买方投资者打造，他们深知，优秀的投资不仅仅在于从低质量渠道获取更多信息。

Portrait was built by former buy side investors, and they understand great investing isn't just about having more information from low quality sources.

关键在于以正确的方式组织正确的信息。

It's about having the right information organized the right way.

如果你收听这个节目，你会明白尽职调查涉及很多方面。

And if you listen to the show, you appreciate diligence consists of many things.

深入探究企业的历史，梳理微妙的竞争动态，追踪围绕你投资逻辑的关键信号。而以往，这会耗费你大量宝贵的时间。

Diving into the history of a business, framing the nuance competitive dynamics, tracking key signposts around your thesis, And historically, that would take up material time that you do not have.

但Portrait就像给你的团队增加了一支分析师大军。

But Portrait is basically like adding an army of analysts to your team.

它由专为投资研究工作流设计的人工智能系统驱动。

It's powered by an AI system specifically designed for investment research workflows.

因此，你能获得细致入微的投资创意。

So you get nuanced idea generation.

Portrait能评估我们节目中讨论的那些定性特征，帮你找到符合自己框架的企业。

Portrait assesses the same types of qualitative attributes that we discuss on this show, and that can help identify businesses which fit your frameworks.

Portrait还能定制研究报告的生成。

Portrait also customizes research report generation.

第三，它还有智能化的投资论点监控功能。

And third, there's intelligent thesis monitoring.

这个功能每天会评估价值链上的数千个数据点，提取关键信息，推动业务发展。

And that's where Portrait assesses thousands of data points across value chains each day, extracting the insights, driving the business.

再说一遍，这些工作通常需要花上好几个小时。

Again, all this work would typically take hours and hours and hours.

现在你随时都能用上它。

It's at your fingertips now.

今天就去portraitresearch.com开始免费试用吧。

Visit portraitresearch.com to start your free trial today.

这里是《商业拆解》节目。

This is Business Breakdowns.

《商业拆解》是一系列与投资者和运营者的对话，深入剖析一家企业。

Business Breakdowns is a series of conversations with investors and operators diving deep into a single business.

针对每家企业，我们会探讨它的历史、商业模式、竞争优势，以及它成功的关键。

For each business, we explore its history, its business model, its competitive advantages, and what makes it tick.

我们相信，每家企业都有值得投资者和运营者学习的经验和秘诀，而我们的任务就是把这些分享给你。

We believe every business has lessons and secrets that investors and operators can learn from, and we are here to bring them to you.

想收听更多《商业拆解》节目，请访问 joincolossus.com。

To find more episodes of breakdowns, check out joincolossus.com.

主持人和嘉宾在节目中表达的观点仅代表他们个人。

All opinions expressed by hosts and podcast guests are solely their own opinions.

主持人、嘉宾、他们的雇主或关联方可能持有本播客讨论的证券头寸。

Hosts, podcast guests, their employers, or affiliates may maintain positions in the securities discussed in this podcast.

本播客仅供信息参考，不应作为投资决策的依据。

This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions.

今天，我们来拆解 Applied Intuition 这家公司。

Today, we are breaking down Applied Intuition.

今天的嘉宾是联合创始人卡萨尔·尤尼斯和彼得·路德维希，他们在2017年创立了这家公司，使命是让十亿台机器变得智能。

Our guests are cofounders Qasar Younis and Peter Ludwig, who started the company in 2017 with a mission to make a billion machines intelligent.

理解Applied Intuition最简单的方式是，它为机器打造大脑，同时也提供工具，让其他公司能自己构建这些大脑。

The simplest way to understand Applied Intuition is that it builds the brains for machines and the tools other companies use to build those brains.

如果一家制造商想让自己的拖拉机、卡车或采矿车实现自动驾驶，它可以直接从Applied Intuition购买智能系统，或者利用其平台自行开发。

If a manufacturer wants its tractor, truck, or mining vehicle to drive itself, it can buy the intelligence from Applied Intuition or use its platform to develop its own.

就像英伟达向其他公司的机器销售芯片一样，Applied Intuition向汽车、国防、采矿、农业和机器人领域的各类机器销售智能系统，而自己并不生产任何一台机器。

Just as Nvidia sells chips into everyone else's machines, Applied Intuition sells intelligence into everyone else's machines across automotive, defense, mining, agriculture, and robotics without building any single machine itself.

我们讨论了为什么未来二十五年最重要的公司都将是实体人工智能公司，他们新推出的代理平台Dana如何用于开发和部署这些系统，以及这家公司如何在没花一分钱的情况下融到了十亿美元。

We discuss why the most important companies of the next twenty five years will all be physical AI companies, Dana, their new agentic platform for developing and deploying these systems, and how the company raised a billion dollars without spending any of it.

请享受这期对Applied Intuition的深度解析。

Please enjoy this breakdown of Applied Intuition.

我知道今天要讲的故事很大程度上会围绕一家公司——Applied Intuition。

I know a lot of the story we're gonna tell today is gonna be about a single business, Applied Intuition.

但这其实也是实体人工智能市场以及自动驾驶技术发展到何种程度的故事。

But it's also really the story of the physical AI market and how far autonomous technology has come.

你们两位在众多行业里都看得最清楚。

And you two see this across as many industries as about anyone.

能不能描述一下物理AI市场的现状，2026年整个格局在你们看来是怎样的，以及从2017年公司成立以来，时间线上有哪些重要的里程碑？

Maybe just describe the state of the physical AI market, how the whole landscape feels to you now in 2026, and maybe some of the important key hash marks on the timeline since when you started the company in 2017.

对我们来说，对Applied Intuition来说，我们的使命是让十亿台机器变得智能。

In our case, in Applied Intuition's case, our mission is to make a billion machines intelligent.

一个简单的理解方式就是自动驾驶汽车。

One simple way of that you could think of is self driving cars.

那些是智能机器，但这只是一个例子，就像Instagram是手机上的一个应用一样。

Those are intelligent machines, but it's one example, like Instagram is an app on the phone.

还有很多其他的应用。

There's many also other apps.

所以物理AI通常是AI与硬件的结合，但发生在现实世界中。

So physical AI is this intersection of AI and hardware typically, but in the real world.

人形机器人也属于这个类别。

Humanoids falls into this as well as a category.

物理AI的技术挑战和数字AI很不一样，数字AI就像大语言模型、信息检索系统，比如聊天机器人这类东西。

The particular technical challenges of physical AI are quite different from digital AI, which is like your LLMs and your information retrieval systems, like chatbots, stuff like that.

因为你要面对现实世界的各种限制，还要考虑物理世界的安全关键性。

Because you have the constraints of the real world, you have the safety criticality of the physical world.

通常我们讨论移动机器时，它们是在有人的时间和空间里移动，这就变成了一个必须认真考虑安全的问题。

Often when we're talking about moving machines, they're moving in a time and space with humans, and suddenly that becomes something where you have to really think about safety.

还有问题的实时性。

The real time nature of the problem.

所以如果你问聊天机器人，给我讲讲彼得·路德维希，它可能花二十秒来处理这个信息。

So if you ask the chatbot, tell me about Peter Ludwig, it can take twenty seconds to process that information.

但当你高速行驶在公路上，一个人形机器人正在做决策，那里就有非常严格的实时性要求。

But when you're flying down the highway, a humanoid is making a decision, there's very hard real time constraints there.

另外，物理AI中可能被低估的一个方面是成本。

And then probably an underreported aspect of physical AI is the dollars.

你需要把这些技术放到机器和芯片上，而且这些机器和芯片在你所讨论的应用场景中必须负担得起。

You need to put this on machines and on silicon that is affordable within the use case that you're talking about.

你不能随便拿无限算力去处理问题。

You don't just throw endless compute at processing something.

你得在一定的算力范围内完成，不仅有时间限制，还有成本限制。

You have to do it in a compute envelope, not only a time envelope, but also like a cost envelope.

这也有助于

It's useful also

区分数字AI和物理AI。

think about the separation of digital AI and physical AI.

对吧？

Right?

数字AI通常就是你电脑或手机上用的那种，屏幕上显示AI处理的结果。

So digital AI typically thinking about what you're using on your desktop or your mobile phone, where there's a screen that's showing you the result of the AI.

而一旦涉及到现实世界中真正在移动的东西，那就进入了物理AI的范畴。

And then where that crosses into physical AI is when anything is, in the real world, actually moving.

这才是真正会对经济产生影响的领域，因为所有涉及移动物体的行业都会因此改变。

And this is where the real impact on the economy will happen when you talk about all of the industries that fundamentally have moving things.

想想看，从工业公司、制造业，到医疗和能源领域的应用，范围非常广。

So think about anything from industrial companies, manufacturing, use cases in health care and energy.

有太多不同的领域，要想真正享受到AI的好处，就必须去影响这些物理系统。

There's so many different fields where in order to get the benefits of AI, you actually have to impact these physical systems.

我再补充一点，通过给机器装上自动驾驶和智能系统，我们其实是在让地球上一些最辛苦的工作变得轻松。

Just to add on to that, by putting self driving and intelligence on machines, we're really making some of the worst jobs on the planet easier.

在数字AI领域，大家总是焦虑不安，担心会计师甚至播客主持人会失业。

In digital AI, there's a lot of teeth mashing and hand wringing about what's going to happen to accountants and maybe even podcast hosts.

但在物理AI的世界里，情况就完全不同了。

You know, in the physical AI world, it's very, very different.

大家恨不得AI能立刻到位。

Like, the AI can't get there fast enough.

比如农业，美国农民的平均年龄已经58岁了。

If you look at, like, farming, the average American farmer is 58 years old.

长途货运行业也面临着创纪录的司机短缺。

We have record shortages in long haul trucking.

比如采矿，全球劳动力中只有1%从事这个行业，但却占了工作相关死亡人数的8%。

Mining, as an example, is 1% of the world's workforce, but accounts for eight percent of work related fatalities.

你想想看，为什么大家都不愿意去矿上工作？

I mean, you can peel that back, like, why do people not want to work on mines?

这些工作太辛苦了。

These are difficult jobs.

人们就是不想在危险的环境里远离家人。

People are just choosing, they don't want to be away from family in unsafe circumstances.

在需求旺盛的领域，影响可以非常快、非常显著。

The impact can be very, very quick and very, very significant in areas where there's a lot of demand.

所以我觉得这是需要记住的重要一点，物理AI和数字AI在这方面真的很不一样。

And so I think that's one of the important things to keep in mind, which is very different in physical AI than, frankly, digital AI.

我觉得二十五年后回头看我们现在这个商业和技术的阶段，我们正在经历很多非常快速的变化。

I think when we look back twenty five years from now about this particular state of business and technology phase that we're going through, we're really going through a lot of changes very, very quickly.

我猜我们不会只讨论代码完整的产品。

My hunch is I don't think we're gonna only be talking about Code Complete products.

这些都很重要，而且你能看到，仅仅这一个应用场景就能产生多大的影响，这也能反映出人工智能对整个社会能带来多大的改变。

Those are important, and you can kind of see how big of an impact just that use case is making as a proxy to how big of an impact that AI can have on society.

我觉得二十五年后回头看，那些做物理人工智能的公司会成为大家心目中真正的主导者。

I think when you look back twenty five years from now, I think physical AI companies are gonna be the ones that really dominate everyone's mind.

卡萨，我知道你一直认为物理人工智能的市场规模会比数字人工智能大好几个数量级。

And, Qasar, I know you have this whole notion that the physical AI market is gonna be in orders of magnitude much larger than the digital AI market.

不如你展开说说为什么这么想，让我们更好地理解这个市场。

Maybe just unpack why you think that and just help us appreciate this market.

我们来看看工业领域这个类别。

Look at industrials as a category.

工业是经济中一个很大的板块，大概占GDP的5%左右。

A huge category of the economy is, you know, roughly 5% of GDP.

其中汽车工业是工业里最大的部分，占全球GDP的3%，这个数字高得惊人。

Automotive of that is the largest of the industrials, which is 3% of global GDP, which is an astronomically high number.

所以当汽车——也就是个人拥有的乘用车——变得智能化时，对每天与汽车打交道的人，也就是地球上几乎所有人，产生的影响会非常非常大。

And so when automotive, which is personally owned passenger vehicles, when they become intelligent, the impact to all the people that interact with cars every day, which is essentially everybody on the planet, is really, really big.

如果你坐在机场里，环顾一下登机口，有多少人跟汽车打过交道，又有多少人那天写了软件？

If you're sitting in an airport and you look around the gate, how many people have interacted with a car versus how many people wrote software that day?

那么，我们再看看其他领域。

So, then you go into the, let's say, other verticals.

Applied Intuition 在这些领域都有涉足。

So, Applied Intuition, we play in all of these verticals.

我们是全球唯一一家做这件事的公司。

We are the only company on the planet that does this.

这包括中国的生态系统和其他地区。

This includes the Chinese ecosystem and others.

我们涉足的其他领域包括商用卡车、国防、建筑、采矿、农业和机器人。

So the other verticals that we're in are commercial trucking, defense, construction, mining, agriculture, robotics.

这些领域每一个本身都非常庞大，无论你用什么标准来衡量。

And each of those verticals is very, very large in itself, in whichever way you want to define it.

无论是纯粹的GDP数字、增长率，还是这些行业的就业人数，几乎在每一个可衡量的方面都是如此。

In just pure GDP numbers, growth numbers, number of people employed in those sectors, almost in every measurable way.

有时候这些市场规模大到让人难以想象。

Sometimes the markets get so big, they're almost like hard for folks to put their head around.

不过，就拿自动驾驶出租车来说吧，它是自动驾驶汽车的一个具体应用，而Waymo这家公司，被相当精明的投资者估值到了1260亿美元。

But I mean, you can just use like robotaxi as one instantiation of self driving cars, and you have Waymo being valued at a 126,000,000,000 by fairly sophisticated investors.

他们给Waymo估值1260亿美元，可不是为了图个虚高的数字。

They're not valuing Waymo at $1.26 because they just want a high valuation.

而是因为它的潜在影响非常巨大，而这还只是其中一个市场的一部分。

It's because the impact can be very, very large, and that's one part of one of those markets.

在我们深入之前，我觉得Applied Intuition这家公司，几乎有种Palantir那种神秘感，就是很难说清楚你们到底是做什么的。

And before we go any deeper, Applied Intuition, in my sense, almost has this Palantir mystique around it, where it's actually very hard to describe what it is and what you guys do.

所以，或许先给我们一个定位，用最简单的话描述一下，你们到底研发和销售什么产品。

So maybe just to orient us, literally describe as simply as you can, what you guys actually build and sell.

Applied Intuition是一家物理人工智能公司。

Applied Intuition is a physical AI company.

我们把智能，也就是人工智能，部署到机器上，让这些机器变得更聪明。

We take intelligence, so we take AI, and we put it on machines and make those machines smarter.

所以不管是让它们自己驾驶，还是你能像跟手机互动那样跟它们进行智能交流，但同时又结合了现实世界的场景。

So whether it's having them drive themselves or whether your ability to interact with them and have intelligent interactions just like you would with your phone, but with the context of the real world.

所以会用到更多的传感器，处理更多的信息。

So a lot more sensors, a lot more information.

然后我们会做所有你能想到的、为了实现这个目标该做的事情。

And then we do all the stuff that you would think in order to get there.

比如我们开发自己的模型，训练和部署这些模型，评估它们，确保它们的安全性。

So we develop our own models, train and deploy those models, we evaluate those models, we make sure that they're safe.

这算是公司的一半业务吧。

That's one, let's say, half of the company.

我们把模型部署到机器上。

We're putting models on machines.

但这比我刚才说的要复杂得多，因为机器的种类非常多样化。

And that's way more complex than even what I just said because the machines have a huge diversity in them.

作为开发者，当你把东西放到手机或笔记本电脑上时，有很好的操作系统来屏蔽硬件细节，让软件不用管硬件的事。

When you're putting something on a phone or a laptop as a developer, you have these great operating systems that abstract hardware and away from software.

如果你是在为联合收割机写软件，那和为汽车写软件完全不同，而汽车软件又和为人形机器人写软件大不一样。

Well, if you're writing software for a combine, that's very different than writing software for a car, which is very different than writing software for a humanoid.

近十年来，我们在这方面做了很多艰苦的工作。

We've done a lot of that hard work in nearly ten years.

公司成立差不多十年了，一直在把这些不同类型的硬件从软件中抽象出来。

The company's almost ten years now in abstracting away all these different types of hardware from the software.

彼得是安卓汽车系统早期的工程师之一。

Peter was one of the early engineers on Android Automotive.

安卓系统在这方面基本上是出了名的。

Android is basically known for this.

它运行在成千上万种不同的设备上，而且是同一个操作系统。

It runs on thousands of different devices, and it's the same operating system.

我们做的也差不多，只不过我们是在上面加上了智能。

We kind of do that except we're doing it with the intelligence on top.

公司的另一半业务则是所有离车软件，也就是用来开发那些智能的开发工具。

And then the other half of the company is all the off board software, the development tools that you would use to develop that intelligence.

所以我们是一家B2B公司，也就是面向企业的公司。

And so we're a b to b company in the sense that we're an enterprise company.

我们向其他企业销售产品，企业通过这两种方式之一找到我们，有时两种方式都会用到。

We sell to other enterprises, And enterprises meet us in one of those two ways, and sometimes in both ways.

他们要么从我们这里购买开发环境，用来打造自己的智能机器，要么直接购买模型，然后直接部署到机器上。

They're buying the development environment from us so they can make their own intelligent machines, or they just buy the models and they put them directly on the machine that way.

我们是一家技术提供商。

It's a technology provider.

我们实际上有两大产品板块。

We really are two big product areas.

我们有外派工程师——我们叫他们别的名字——但绝大多数员工，我们更像是一家产品公司。

We have forward deployed engineers, we call them something else, but most vast majority of our workers, we're like a product company.

从这个角度看，我们其实更像一家芯片公司。

In that way, we're much more actually like a silicon company.

就是说，我们是技术提供商。

Like, we're a technology provider.

这有点像你想到芯片时的感觉。

It's almost like you think about chips.

芯片可以装进各种不同的机器里，做各种不同的事情，但它们本质上是一个平台。

They go into all these different machines, and they can do all these different things, but they're kind of a platform.

我们跟这个有点像，只不过我们的平台不是硅。

And we're kinda like that except our platform isn't silicon.

而是智能。

It's intelligence.

想象一下，你经营一家制造某种机器的公司。

Imagine that you run a company that makes some kind of machine.

这个机器可能用在交通领域，或者机器人领域，又或者医疗健康领域。

Again, this this maybe is in transportation, something in robotics, or something in in in health care.

你想让这台机器变得智能。

And you wanna make this machine intelligent.

对吧？

Right?

假设这台机器有传感器和执行器，你想让它智能地完成某些任务。

So the machine, let's say, has sensors and actuators and you wanted to do something intelligently.

那具体要怎么做呢？

How do you actually do that?

如果你想自己开发这项技术，就需要一个非常强大的工具平台来支持开发工作。

Well, if you want to develop the technology yourself, you're going to need a really strong tooling platform to actually do that development.

所以Applied Intuition做的就是开发和销售这种工具平台，供制造那台机器的公司的工程师使用。

And so Applied Intuition, we we make and we sell that that tooling platform used by engineers at the company that's building that machine.

或者，作为机器制造商，你可能更想购买一个更完整的解决方案，让机器几乎开箱即用就能变得智能。

Or maybe as the maker of the machine, you actually want to purchase more of a complete solution that can make the machine intelligent almost out of the box.

我们也会制作这类更完整的解决方案，然后同样销售给这些公司。

We also make more of those complete solutions, which we then sell to those companies as well.

我们的产品范围从工具到解决方案都有，然后把这些技术授权给行业使用。

We have that spectrum from tools to solution, then we license this technology out to the industry.

我上周在旧金山，基本上每两辆车里就有一辆是Waymo或者其他自动驾驶汽车。

I was in San Francisco last week, and, basically, every other car now is a Waymo or some other autonomous vehicle.

看到这项技术像爆炸式地发展，感觉挺酷的。

And it's kind of cool to see the, like, explosive nature of that technology.

但还有很多其他技术，比如无人机、人形机器人、机器人技术、采矿技术、农业技术等等，如果我们能拥有它们，那将非常棒，而且我们立刻就能看到它们潜在的巨大价值。

But there are also a bunch of other technologies, whether it's drones or humanoids, robotics, mining technology, farming technology, etcetera, that if we had them, it would be amazing, and we could immediately see how valuable the potential would be.

但很难预测这些技术在实际时间线上会如何发展，而你已经在很多不同的技术领域开发了工具。

But it's very hard to actually predict how long in the timelines these things play out on, and you've been able to develop tools across a bunch of these different technologies.

我很好奇，你是如何保持灵活性，为这些难以预测的技术构建工具和系统的？

I'm curious how you were able to stay flexible and be able to build tools and systems for these technologies which are hard to predict out.

在创立Applied Intuition之前，我是Y Combinator的首席运营官。

Before Applied Intuition, I was the COO at Y Combinator.

Sam Altman是总裁。

Sam Altman was the president.

我是首席运营官。

I was the COO.

我负责管理这家公司。

I ran the firm.

那段时间正是OpenAI创立的时候，也是我们投资Cruise、Scale以及其他一批如今在这个领域很出色的公司的时候。

It was the era of where OpenAI is created, and it's the era where we fund Cruise and Scale, and a bunch of other great companies that are kind of in this space now.

我讲这个背景是因为，对正在听的创业者来说，最重要的一点就是时机决定一切。

I give that context because the most important thing, especially for founders who are listening, is timing is everything.

如果你开发的技术早了两三年，市场还没准备好接受它，你就会花很多钱等着市场成熟——我觉得这其实是大多数公司失败的常见原因。

If you build a technology that's maybe two years too early, the market is not ready to consume it, and you burn a lot of money waiting for the market to mature, which, by way, I think is the default failure case.

大多数公司失败是因为起步太早。

Most companies fail because they're too early.

很少是因为起步太晚而失败的。

Rarely do they fail because they're too late.

但反过来，太晚也不行，因为竞争会非常激烈。

The opposite, though, is you can also be too late where it's just very competitive.

玩家太多了。

There's lots of players.

市场的利润空间基本被挤压没了。

The margins in the market is kind of being destroyed.

我虽然是个工程师，但也在哈佛商学院读过研究生。

Know, I I'm an engineer, but I also did a grad degree at HBS.

我有个MBA学位。

I'm an MBA.

市场动态这方面，有时候确实被低估了。

That aspect of market dynamics is sometimes under emphasized as well.

那么，带着这个背景，差不多十年前我们是怎么应对的呢？

So, with that context, how did we navigate nearly a decade ago?

我和彼得在创立这家公司时，采取了极其审慎的方式。

Peter and I took making this company in an extremely intentional way.

我们不是靠瞎猜走到这一步的，毕竟我们年纪也不小了。

We didn't just guess our way into a part of it because we're old.

我们之前做过公司，也在谷歌和其他地方领导过大型工程团队。

So we've done companies before, and we'd led large engineering teams at Google and other places.

我们当然有技术经验，但其中一部分也是因为我们非常刻意地给自己设定了这些限制条件。

We have, you know, technical experience, but part of it is also just being very, very intentional about putting these constraints on.

所以最初，我们第一次讨论一起创业时，甚至在我加入YC之前，还在谷歌共事的时候，我们就想，也许该搞一家自动驾驶出租车公司。

And so initially, what we envisioned was the first first time we talked about working in a company together, even before I was at YC, when we were working together at Google, was, hey, we should maybe do a robotaxi company.

当时我们得出的结论是，那会儿是2010年代初，技术还没成熟，这意味着你会太早入场，得花大把时间等技术变成能量产的产品。

We concluded at that time, this is in the early teens, that it's simultaneously the technology hasn't really been figured out, which means you're going to be too early, you're going spend a lot of money waiting for the technology to manifest itself into a production product.

其次，商业模式也没成型。

And then secondly, the business model hasn't been figured out.

比如，自动驾驶出租车到底怎么才能盈利？

Like, how is the robotaxi gonna be really a profitable venture?

后来我去了Y Combinator，彼得留在了谷歌。

I ended up at Y Combinator, Peter stayed at Google.

再后来，我们在YC投资了Cruise，然后Cruise被通用汽车收购了。

And then fast forward, we funded Cruise at Y Combinator, Cruise was bought by General Motors.

我本科是在通用汽车学院读的。

So I went undergrad at the General Motors Institute.

我也在通用汽车工作过。

I worked at General Motors.

彼得的父亲和祖父都在通用汽车工作过。

Peter's father, grandfather worked at General Motors.

所以我们俩都是底特律出身的人。

So we're we're both Detroit guys.

我们的家族根基深深扎在汽车行业里。

Our family roots are very deeply in the automotive industry.

2016年，当Cruise被收购时，收购方正是通用汽车，我们又开始讨论这个行业正在发生什么。

In 2016, when Cruise was acquired, it was acquired by none other than General Motors, we started talking again about what's happening in this industry.

当时这个行业特指汽车行业。

This industry at the time specifically was automotive.

汽车行业往哪儿走，说实话，国防工业就往哪儿走，建筑和采矿行业也是，农业也一样。

Where automotive goes, that's where honestly defense goes, and that's where construction and mining goes, and that's where agriculture goes.

因为制造运输系统的方式——比如卡特彼勒、小松或者约翰迪尔这样的联合收割机——其实和汽车算是近亲产品。

Because the ways that you build a haul system, you know, if you're if you're a Caterpillar or Komatsu or a combine like John Deere, is actually kind of like a cousin product to a car.

或者像通用动力公司，他们制造步兵班用车、部队运输车之类的，也是类似的情况。

Or if you're General Dynamics and you're building an infantry squad vehicle, a troop mover, or something like this.

我们当时就想，好吧，这个行业会往哪个方向发展呢？

When we said, okay, well, where is this industry going?

这个行业指的就是汽车行业。

This industry being automotive.

我们觉得，这个行业会经历类似特斯拉化的变革。

We're like, well, there's gonna be kind of like the Tesla fication of this industry.

这些机器会变得智能。

These machines are gonna get smart.

它们会以软件为核心。

They're gonna be software first.

然后会有各种工具来实现这些功能，比如车队管理、软件更新，甚至测试软件以确保其可靠性。

And then you're gonna have all these tools that are going to actually enable that to happen for fleet management, to updating software, to literally testing the software to make sure it's dependable.

这是基本原则，我这么说也是给那些想自己创业的人做个梳理。

First principle, so I'm just kind of enumerating this for people who are going to start companies themselves.

我们当时想，如果现在做软件并试图卖给制造商，他们不会从一家小公司那里采购，因为这些都是涉及安全的关键系统。

We thought, okay, well, if we make software right now and try to sell it to the manufacturers, they are not going to consume it from a little young company because they're safety critical systems.

你需要更多的业绩记录和实力，坦白说，这些系统非常复杂。

You need a lot more track record and heft, and they're frankly very complex systems.

一个50人的团队是造不出自动驾驶汽车的。

A team of like 50 people cannot build a autonomous vehicle.

子系统太多，复杂度太高了。

There's too many subcomponents and complexities.

所以我们从工具入手。

And so we started with tools.

今天我们要重点聊的是公司在这个核心品类里最大的一次产品发布，叫做Dana，它是一个智能代理平台，用来完成我们过去近十年一直在做的事情，但这次是以更AI优先的方式来做。

And one thing we're here to talk about really is our biggest product launch in that fundamental category of the company, which is a product called Dana, which is an agentic platform in order to do everything we've been doing for the last almost ten years, but doing it in a much more AI first way.

不过，我们就是这么起步的。

But that's kind of how we started.

既然你提到了数据，给大家补充一点背景信息。

Since you brought up data, helpful context for everyone.

如果回顾一下我们业务的发展历程，你提到你们是从工具起步的，而不是走垂直整合路线去造自动驾驶汽车。

If we trace the evolution of the business, you mentioned that you started with tools instead of the vertical integrator route, building autonomous vehicles.

所以你们从工具起步，构建了操作系统，现在又有了自动驾驶堆栈，还有了Dana。

So you start with tools, you build the OS, now you have the Autonomy Stack, and now you have Dana.

彼得，或许你可以帮我们回顾一下这个演变的历史，以及这些部分是如何协同运作的。

Peter, maybe it's helpful for you to walk us through the history of that evolution and how it all plays together.

首先，我想说的是，大约十年前我们刚开始时就明白，技术还会发生很大的变化。

Firstly, I would say something that we we knew when we started almost ten years ago was just that the technology is still going to change a lot.

整个领域有大量的先进工程和研究在推进。

There's a lot of advanced engineering and research that works in this entire field.

要想参与其中，但又不过度依赖某个具体实现方式，更好的思路是横向思考。

A way that you can be part of that, but not be, let's say, overly exposed to any specific implementation, is to think more horizontally.

对我们来说，这意味着最初专注于工具，并以一种可以不断往平台上添加功能的方式来构建工具，同时认识到，当我们谈论物理AI和高级自主系统时，几乎每两年就会出现某种突破，改变我们对这些问题的思考方式。

And so for us, that meant initially really focusing on tools and then building tools in such a way that we could continue adding onto the platform and just recognizing that the technology itself, when we talk about physical AI and advanced autonomous systems, almost every two years, there's some sort of breakthrough that changes how you have to think about these things.

如果你不够灵活，不能及时适应最新的技术或突破，那么你就有可能变得过时。

If you're not dynamic enough in in understanding how to adapt that that latest technique, that latest breakthrough, then you can become almost obsolete in that sense.

这一点已经深深融入了Applied的基因中。

And that that's been really baked into Applied's DNA.

这几乎就像是我们必须对自己进行的一种内部颠覆，以确保我们始终处于领先地位。

It's almost like this internal disruption that we sort of have to do to ourselves to make sure that we stay on top of things.

所以，工具最初是很好的方式，可以横向地应用于所有这些行业。

And so tools was was a great way of doing that initially and doing that horizontally across all these industries.

在工具开发几年后，你会发现，当把这些技术部署到机器上时，问题远不止工具本身那么简单。

What happened after a few years of working on tools is you hit a point deploying this technology onto machines that the problem is much larger than the just the tools.

于是你不得不思考，瓶颈到底在哪里？

And you have to think, well, what are the bottlenecks?

哪些因素是限制速度的关键？

What what are the rate limiting factors?

再次强调，我们的北极星是希望产生重大影响。

Again, sort of with this this north star of we wanna have a big impact.

我们希望让十亿台机器实现自主化。

We wanna make a billion machines autonomous.

我们希望安全高效地做到这一点。

We wanna do this safely and efficiently.

你会突然发现，操作系统本身反而成了瓶颈，尤其是在把这项技术部署到车辆上的时候。

You hit this point where all of a sudden, the operating system actually becomes a bottleneck, like deploying this technology onto the vehicle itself.

这几乎是被迫进入了那个领域，我们不得不

It sort of forced us into that business, and we had

去构建一个真正

to build a a really

优秀的解决方案，让你能把软件部署到机器上，可靠地更新软件，并拥有所有正确的诊断功能。

good solution that allows you to deploy software onto machines, to update that software reliably, to have all the right diagnostics.

在机器上运行高级模型和神经网络，是极其复杂的。

And running advanced models, neural networks on machines, is extremely complicated.

有时候人们会把它简单化，觉得哦，不就是模型的事嘛。

Sometimes it gets trivialized, and people think about, oh, it's just about the model.

但实际上，要让这一切运转起来，你需要解决上千个不同的问题。

But really, there there's about a thousand different problems you have to solve to make this all work.

而操作系统这部分在其中占了很大比重，所以我们不得不去解决它。

And the operating system piece is is a really big part of that, and so we had to solve that.

当我们有了这两个部分，也就是工具平台和操作系统平台之后，我们就得开始考虑打造更完整的解决方案了。

And once we had those two components, right, we had the the tooling platform, we also had the operating system platform, then we have to start thinking about creating more of that full solution.

这也正是我们真正进入垂直自主技术栈的原因，我们自己开发更多模型，并把这些模型提供给客户。

And that's what brought us really into the vertical autonomy stack, doing more of these models ourselves and making those available to customers.

现在，我们在很多领域都拥有非常完整且极具吸引力的产品方案。

And now we have a really complete and very compelling offering in a lot of these areas.

对。

Yeah.

我觉得在这个概念里，还有一点值得强调，就是横向公司与纵向公司的区别。

I think also worth highlighting in this concept of, like, a horizontal company versus a vertical company.

纵向公司很好理解，比如特斯拉就是一家纵向公司。

The vertical companies, it's kind of easy to understand, like a Tesla is a vertical company.

而横向公司就像英伟达那样。

A horizontal company is like an Nvidia.

这种公司把技术卖给广泛的客户，客户再把这些技术整合起来推向市场。

It's a company that sells this technology across a broad base of customers who then package it together into something and take it to market.

重要的是，我们为一个垂直领域打造的东西，也能在其他垂直领域用得上。

It was important that the stuff that we built for one vertical could be used in other verticals.

所以我们还加上了这个限制条件。

And so we also put that constraint on.

所以再说一遍，给在家创业的创始人提个醒：光有创办公司的雄心是不够的。

So again, for the founders at home, it's not enough that you have, like, ambition in building a company.

你的想法还得是正确的。

Your ideas also have to be, like, correct.

大概十年前，大家总爱问：为什么工具类业务不好做？

Almost like 10 ago, the question always was like, Well, why tools is like a bad business?

为什么要做横向平台？

And why be horizontal?

垂直才是正确答案。

Vertical is the right answer.

我真的希望大家都能从第一性原理出发去思考，我们的成果已经说明了一切。

You know, I really implore everybody to really think from first principles, and our results speak for ourselves.

我把这归功于我们的技术策略，就是把我们在汽车领域做的东西应用到国防领域。

And I attribute that to our technical strategy of putting the stuff that we make in automotive onto defense.

把国防领域做的东西应用到建筑和采矿等领域，以此类推。

Putting the stuff we make in defense onto construction and mining and so on.

我知道你们有一个宏大的愿景，你们也提到过，就是未来十年要实现十亿台智能机器。

I know you guys have this, like, grand vision, and you guys mentioned it, of getting to a billion intelligent machines over the next decade.

我们今天要聊的是Dana，以及它如何实现并开启这种可能性。

We're here to talk about Dana and how that's gonna enable and unlock that possibility.

不如就说说Dana是什么，以及它将如何帮助我们走向未来。

Maybe just talk about what Dana is and how that's gonna help us get to the future.

好的。

Yeah.

Dana是我们为物理AI打造的新一代智能平台。

So Dana is our new agentic platform for physical AI.

这实际上是我们过去十年所有工作的结晶，它让开发这些系统变得比以往容易得多。

This really is the culmination of pretty much everything we've worked on now over the last decade, and it makes developing these systems so much easier than it has been in the past.

不过，一开始就得明白为什么这很重要。

It's important to understand why that's important at the outset, though.

构建物理人工智能极其复杂。

So building Physical AI is extremely complicated.

如果你问，为什么现在还没有到处都是智能机器人和智能车辆？

If you ask, why don't we have intelligent robots and intelligent vehicles everywhere today?

归根结底，是因为开发这些东西真的很难。

It really comes down to building this stuff is really hard.

这就是其中的限制因素。

That is the limiting factor in it.

我们知道如何制造芯片。

We know how to make the chips.

我们知道如何为这些系统制造硬件。

We know how to make the hardware for these systems.

更关键的是，实际开发所有这些技术并让它们运转起来，非常非常困难。

It's more so actually developing all this technology and getting it to work is very, very difficult.

过去十年我们的工程工具，确实解决了一部分问题。

Our engineering tools over the last ten years, they they are addressing parts of this.

但现在，借助现代人工智能和这个新数据平台，我们正在大幅降低构建物理人工智能并将其部署到现实世界的门槛。

But now with modern AI and with this new data platform, we're we're really drastically reducing the barrier to entry to building physical AI and deploying it in the real world.

随着门槛的降低，我认为这将让构建各种解决方案变得容易得多，真正为我们的客户赋能。

With that lowering of the barrier to entry, I think it's going to make it far easier to build a very large variety of solutions and really supercharge our customers.

举个例子吧。

Use an example.

如果你现在想给iPhone开发一个应用，高中生都能做到。

If you're building an app for an iPhone, like, high school kids can do that now.

因为现在有各种现成的工具，让这件事变得很简单。

Because there's all of these things that exist and and and that make it easy.

而且有了新的编程平台，比如Vibe编程平台，开发一个网页应用比以往任何时候都更容易，超级简单。

And with the new coding platforms, like Vibe coding platforms, it's easier than ever to make a web app super, super simple.

但在机器人领域，要做到这一点就非常困难了。

It's very hard to do that in terms of robotics.

如果你想做一个大学校园里的送货机器人，或者一个能帮你打扫屋子的小吸尘器，那真是件挺吓人的事，就算对爱好者和计算机科学家来说也一样。你得把很多很多互不相关的产品和工具拼凑在一起，然后还得想办法把软件部署到那台实体机器上。

If you wanted to build like a delivery robot for college campuses, or like a little vacuum that cleans your house, it's a pretty daunting thing, even for hobbyists and computer scientists, like, you have to patch together lots and lots of disparate products and tools, and then you have to somehow figure out how to deploy that software onto that physical machine.

Dana 正是这样的平台，而且它还把很多那种智能代理能力带到了纯粹为网页应用编写软件这件事上。

Dana really is that, along with the fact that it brings a lot of that agentic power in writing software purely just for, like, web applications.

Peter 一直在强调，也是我们真正想做的，就是降低门槛。

The thing that Peter's really emphasizing, and thing that we really want to do is just lower the bar.

所以，外面任何人——先从工程师开始，但最终，任何人都能开发机器人。

So anybody out there, starting with engineers, but ultimately, anybody can develop robots.

我觉得这真的让我们离那个使命更近了一大步。

And I think that really takes us much, much closer to that mission.

老实说，我觉得几年前这是不可能的，因为我们当时没有那种智能，也就是那些模型，来帮助我们创造数据，然后让终端用户用数据去真正创造智能。

I think, frankly speaking, it wasn't possible a few years ago because we didn't have the intelligence, literally the models, that would help us create data, and then for end users to use data to actually create intelligence.

在打造 Dana 的过程中，有多少动力是源于预见未来的发展方向？

How much of the drive to build Dana was building where the puck is going?

又有多少是源于客户的痛点或摩擦点——我们可以列举一些——然后我们需要为他们去构建解决方案？

And how much of it was these are customer pain points or friction points, and maybe we we can enumerate some of them, and that we need to go ahead and build the solution for them.

这就像是我们正在构建的技术栈的下一个进化阶段。

This is just like the next evolution of the stack that we're building.

对吧？

Right?

工具、操作系统、自动驾驶栈，现在又有了Dana。

Tools, OS, Autonomy Stack, and now we have Dana.

这其中有多少是外部需求驱动的，又有多少是内部想法推动的？

How much of it was, like, outside in versus inside out?

两者都有。

Both.

我说两者都有，是因为我们自己也在用自家工具开发自动驾驶系统。

I say both is we use our own tools to develop Autonomy as well.

所以我们自己就是客户，而且公司内部不同部门的需求也不一样。

So we're our own customer, and those are different parts of the company.

我得说，最苛刻的客户反馈其实来自内部，他们对任何不能快速、高效、一次成功的东西几乎零容忍。

And I would say our most aggressive customer feedback comes from internally, where there's, like, very little patience for for anything that doesn't work quickly and fast and and on first try.

我们是一家企业级公司，近十年来一直在向客户部署工具，并从成千上万依赖我们的工程师那里获得反馈。

We're an enterprise company, and for almost a decade, we've been deploying tools to customers, and we get feedback from many thousands of engineers who depend on us.

他们也在使用像Claude这样的工具。

They're also using things like Claude.

所以很自然，如果你在用Cursor和Claude，就会想，嘿，那个CLOD cursor在物理AI领域对应的是什么？

So then it's very natural if you're using Cursor and Claude and say, hey, where is the CLOD cursor thing but in physical AI?

而我们正是开发这个的公司。

And we're the company to develop that.

说实话，这其实就是我们的看家本领。

I mean, this is just, frankly speaking, our bread and butter.

我觉得如果你身处这个领域，这一点非常明显。

I think if you're in the space, it's very obvious.

它就应该像使用主流编程平台一样简单。

It should be that easy, just like it is to use one of the major coding platforms.

请帮我们理解一下，为什么Applied Intuition在构建这个方面比其他公司更有优势？

Help us understand why Applied Intuition is uniquely positioned to go ahead and build this rather than another company.

或者像你说的，其他通用的AI助手或编程代理也可能做到一部分。

Or as you said, like, there are other generic AI assistants or coding agents that could potentially do some of this.

那Dana到底有什么不同，为什么偏偏是Applied Intuition来做？

What's so different about Dana, and why Applied Intuition specifically?

所以，得先多了解一下这项技术本身。

So it's important to understand a bit more about the technology itself.

像Anthropic或OpenAI推出的通用模型确实很棒，处理通用任务非常有用。

So the general purpose models that come from companies like Anthropic or OpenAI, those are great, and they're very useful for very general purpose tasks.

但当你面对的是安全至关重要的系统，开发的东西直接关系到人命，而且需要极其复杂的开发工具链时，光靠模型是不够的。

But when you're dealing with things that have a very deep safety critical component, things where lives are literally on the line based on on what is being developed, and they require a very complex development tool chain, Just the model is not enough.

那只是整个解决方案的1%。

That's like 1% of the full solution.

数据平台本身才是关键，它是建立在我们过去十年积累的所有成果之上的。

The data platform itself, right, this is building on top of everything that we've built over the past ten years.

所有东西都有API，并且已经重新架构，让AI代理成为核心。

Everything has an API, has been rearchitected to work in a model with AI agents at the forefront.

构建物理AI需要非常复杂的工作流程，过去可能要在20种不同工具之间切换，还得深刻理解这些工具之间精确的信息流动，才能最终达成目标。

These very, very complex workflows to actually build Physical AI, they maybe would require switching between 20 different tools in the past for different tasks and deeply understanding this precise flow of information between all of those things to accomplish your end goal.

所以，要让一个物理AI系统真正运作起来，技术栈的层次非常多。

So to actually make a physical AI system work, there's many, many layers to that technology stack.

举个例子，为什么不能用Claude来构建Linux内核呢？

For example, it's like, why couldn't you use Claude to build the Linux kernel?

因为Linux内核本身是一个非常复杂的技术，经过多年开发，使用了大量不同的工具。

It's like, well, because a Linux kernel is actually a very, very complex piece of technology that's been built out over many years and using many, many other different tools.

物理AI也是同样的道理。

The same thing in Physical AI.

从根本上说，这些技术都非常复杂。

Like, fundamentally, these technologies are very, very complex.

而现在有了Dana，所有这些复杂的工作流程都可以通过一个智能代理界面无缝编排，你只需用日常语言输入指令，就能得到日常语言的回答，还能完成数据科学和生产部署中那些非常细致、深入的任务。

And now with Dana, all of those complex workflows can be very seamlessly orchestrated from an agentic interface, where you're able to write things in plain English and receive answers in plain English, and do very detail oriented things that are very deep in data science and production deployment of this type

这就是AI的体现。

of AI.

举个例子吧。

Use an example.

就拿自动割草机来说。

Let's use a autonomous lawnmower.

你需要哪些东西？

What are all the things that you need?

首先，需要一些传感器。

First, need some sensors.

还需要一些计算设备。

You need some compute.

然后你得做一个软件包，让它知道这是我要工作的院子。

Then you need to create a software package that understands, this is the yard I'm gonna work in.

这是我要工作的物理空间。

This is the physical space I'm gonna work in.

别跑到其他地方去。

And don't go in other places.

所以现在传感器得理解那个物理空间。

So now the sensors have to understand that physical space.

然后你想创建一些场景，让那台割草机在模拟中成功通过。

Then you wanna create scenarios that that lawnmower has to successfully pass in simulation.

所以你需要一个模拟框架。

So you need a simulation framework.

那么怎么模拟你的后院呢？

And so how do you simulate your backyard?

这本身也挺复杂的。

And so that also is complex.

模拟本身就是一个庞大的行业。

Simulation itself is a massive industry.

有些公司市值几百亿美元，只做模拟这一块。所以我们起家的核心业务就是模拟，也就是世界模型这个领域。

There are companies that are worth tens of billions of dollars that only do simulation, so And that's where we started our bread and butter, was in simulation, in the world model universe.

现在你有了一个模拟的后院，好，那你在模拟后院里有了可以反复运行的场景，而且这些必须在云端完成。

So now that you have a simulated backyard Okay, so now you have scenarios in a simulated backyard that you can run again and again, and you have to do that in the cloud.

这需要一套老式的编排流程。

That is an old orchestration that needs to happen.

最终，当你的效率和精度达到一定水平后，就会把第一个版本部署到实体割草机上。

And then ultimately, once you're performing at a certain level of efficiency and fidelity, then you're going to deploy that first version onto the physical lawnmower.

然后肯定会有很多地方出问题。

And then a bunch of things are not going to work.

你得去搞清楚，为什么没成功？

And you have to figure out, well, why didn't it work?

为什么执行机构没有按你预想的方式运作？

Why didn't the actuation happen as you thought it would?

为什么控制系统可能表现不正常？

Why are the control systems maybe not behaving?

所以这就形成了一个完整的反馈循环。

And so then there's a whole loop of feedback.

这些事光靠大语言模型是搞不定的。

You can't do that all in an LLM.