Figure AI CEO：人形机器人是世界上最伟大的事业

Brett，欢迎来到 Sourcery。

Brett, welcome to Sourcery.

谢谢邀请。

Thanks for having me.

我不知道这会是第一部分还是第二部分，但这是一个系列了。

I don't know if this is going to be part one or part two, but this is a series now.

好的，太好了。

Okay. Great.

我们参观了整个工厂，现在我们要坐下来做个访谈。我想我们最近在 Hill and Valley 的很多犀利观点都火了。所以，我觉得从这里开始会很棒。那么，你现在最犀利的观点是什么？

We did the entire tour, and now we're going to do a sit-down interview. I figured we recently went viral for a lot of these hot takes at Hill and Valley. And so, I think it would be great if we just start there. So, what is your hottest take right now?

关于机器人领域？

On in robotics?

是的。

Yeah.

我的犀利观点是，我们花了很多时间在做完全自主和端到端的事情。你刚进来的时候问了我们一些问题，比如“这是远程操控的吗？”我们并没有远程操控这些东西。我认为我们在机器人领域最犀利的观点是，如果不亲自到现场看看，真的很难看清这个领域的真实情况。所以，我希望你今天在这里看到我们做的一切，能有好的体验。我最犀利的观点是，我们就是想让仿人机器人真正工作起来，而现在它们已经在工作了。就这么简单。我们看到机器人在做日常的事情，比如清理客厅、做商业工作。看到这些真的很酷，看到未来几年这将成为现实，真的很酷。

My hot take. I think one thing that we spend a lot of time on is doing things with like fully autonomous and end-to-end. And you asked us a few questions here when we walked in of like, "Is this like teleoperated?" We're not teleoperating this stuff. I think our hot take for robotics is just it's like it's kind of really difficult to see what's really happening in the space without coming on site and really seeing stuff. So, I hope you had a good experience here today seeing everything we're doing. I think the hottest take I have is like we just want humanoid robots to work and like they're working now. And it's pretty simple. Like we're seeing robots do everyday things like we saw it clean up a living room, do commercial work. Like it's just cool to see it. Like it's cool to see that this is going to happen next few years.

这是一个竞争非常激烈的领域，而且越来越激烈。有趣的是，我们采访 Skydio 时，Adam 谈到了无人机周期。无人机有炒作周期之类的。但仿人机器人肯定处于不同的规模和节奏。那么，面对竞争，你感觉如何？

This is a very hotly contested space and it's becoming more and more competitive. It was funny because we did this interview with Skydio and Adam was talking about the drone cycles. There are drone cycles where there's hype cycles and that kind of thing. But humanoid robots are certainly at a different scale and pace than before. So, how does this feel with the competition?

听着，我们的内部目标是如何让这些东西做真正的事情并因此获得报酬？所以我们想了很多关于如何做自主的有用工作。这是我们的标准。我们需要用 AI 模型很好地做到这一点，并且需要在好的硬件上做到。你知道，要成本效益高，我们能大量制造，我们制造很多机器人。你知道，我们认为可能在这个时间点上，我们比全球所有在这个水平上做这件事的大公司都要领先几年，就像我们处于仿人机器人的早期阶段。希望下一步是如何让更多机器人规模化地部署出去？我们想成为第一个做到这一点的。如何让世界上有成千上万的机器人每天运行？这个领域还很早期。我们就像处于仿人机器人大规模进入社会的第一章。是的，我的意思是，我们内部很兴奋，因为它正在起作用。这只是第一章。第二章是让更多机器人出门，让它们在更大规模上工作。在某个时候，我们想真正实现泛化，做人类能做的一切。

Listen, our internal goal is like how do we get these things to do real stuff and get paid for it? And so, we think a lot about how do we do autonomous useful work? That's our bar. And we need to do that with AI models really well and we need to do it on good hardware. It's like, you know, cost-effective, we can make a lot of it, we make a lot of robots. You know, we think probably at this point a few years ahead of every big everybody globally doing this at this level, which is like we're kind of early in the humanoid book. And hopefully the next step is like how do we get more robots out at scale? And we want to be first to be able to do that. How do we get, you know, hundreds and thousands and tens of thousands of robots in the world that run every day? And the space is just early. There's a lot of we're just like we're in the first chapter of humanoids coming into society at scale. And yeah, I mean we're pumped about here internally as it's working. And it's just chapter one. Chapter two is like get more out the door and get them working even at bigger scale. And at some point we want to really be able to generalize to do everything a human can.

你们一年想生产多少？目标是什么？

How much do you want to produce a year? What's the goal?

今年我们会尽可能快地生产数千台机器人。我们正在尽可能快地提升生产线。三月份我们创下了生产纪录，到五月份我们要把这个数字翻三倍。我们会生产数千台机器人。我们基本上已经备好了零部件，正在提升产量。然后从那里开始，我们要生产数万台、数十万台。我们希望能达到年产一百万台。然后我们需要商业进展也跟上。我们有大量的商业需求，我觉得如果机器人都准备好了，我今天就能把它们部署到商业客户那里。所以最大的差距是让机器人准备好进行大规模的自主操作。

We'll make thousands of robots over the here like basically as fast as we can this year. So we're ramping bot queue lines up as fast as we can. We had record production in March and we're going to 3x that by May. And we'll build thousands of robots. We basically have we already have the parts in house to do this and we're ramping up production. And then from there we want to build tens of thousands and hundreds of thousands. And we want to be able to get to a million units a year. And then we need the commercial progress to also match that. We have so much commercial demand it's like it's hard to we have like overwhelmingly amount of we could I think I could put so many robots into commercial customers today if they were all ready. So the big gap here is getting the robots ready to do scale autonomous operations.

所以现在商业化的瓶颈是

So the bottleneck right now for commercializing them is

是拥有足够的机器人，并让它们在大规模下达到人类水平的性能。我们不想做的是，把一千台机器人推向市场，然后每小时出现一千个问题。

It's having enough of them and making them run to a level of human performance at scale. What we don't want to do is we don't want to put a thousand robots out the market and have a thousand problems every single hour.

是的。

Yeah.

这对我们谁都不好。所以去年我们有一小批机器人去了宝马，每天工作。我们连续运行了六个月，每天都运行。那太棒了。我们学到了很多。之后我们重构了整个软件和 AI 系统的商业化方法。这让我们推出了 Helix 2，这是我们内部第二代 AI 模型，几个月前发布的。现在的问题是如何把这些机器人部署到许多不同的客户那里，我们可能在未来 90 天内会宣布很多这方面的消息，并在今年以相当的规模部署到这些客户群体中。假设一切顺利，我们会继续扩大。所以去年我们有机器人在那里，进展顺利，我们学到了很多，今年我们会有更多的机器人进入许多不同的客户。然后六到九个月进展顺利，我们就会疯狂地扩大规模。

That's not good for any of us. So we had a small batch of robots that went out to BMW last year and did work every day. We ran for six months every single day. It was phenomenal. We learned a ton. We refactored our whole approach to how to commercialize the software and AI systems after that. And that kind of led us to Helix 2 which is our second generation AI model internally that we launched a couple months ago. And now it's like how do we put these robots into many different customers that we'll probably announce a lot of this in the next like 90 days and put them out into those groups at like decent scale like this year. And assuming that goes well, we'll just keep compounding that. So we had robots there last year, that went well, we learned a lot, we'll have a much larger amount of robots going into many different customers this year. And then six to nine months goes well, we'll just keep scaling like crazy.

你担心 Optimus 吗？

Are you worried about Optimus?

在我看来，这不是一个制造问题，而是一个智能问题。

In my mind, this is not a manufacturing problem. This is an intelligence problem.

你们公司快四岁了。你们是怎么这么快就扩张起来的？过程是怎样的？

You're almost four years old. How did you scale up this fast? What was the process like?

是的，我创业大概有 20 年了。我把一家软件公司快速做大后卖掉了，又把 Archer 快速做大并成功上市。所以每次处于这个阶段，我都会坐下来反思：我从过去的经历中学到了什么，怎么才能做得更好？我觉得我们采取了一种非常差异化的方法，基本上就是垂直设计一切。我不认为世界上有任何机器人团队会比我们设计更多的零部件。我们设计电机，基本上里面的每个部件——转子、定子，所有东西。传感器、结构、运动学、关节，还有你今天看到的电池和电池组。这真的让我们能够掌控自己的命运。我们可以建立自己的供应链。如果没有这些，你就只能受制于某个供应商，如果它出了问题，你怎么解决？如果是代码问题，你懂吗？你能做质量保证吗？你能修复吗？你能打补丁吗？所以我们从上到下理解整个技术栈。前期要找到能胜任这些工作的人才，付出了巨大的努力。然后我们一直迭代到现在，已经拥有相当可靠的系统，运行得很好。我一开始是自筹资金创办了整个公司。四个月内我们就达到了每月 100 万美元的烧钱速度。这可不是闹着玩的。四五个月内我们就有了一支 40 人的团队，他们都非常优秀。然后就在这里，每周工作 100 小时，和团队一起努力让它运转起来。我们犯过一些错误，也学到了很多。有些事情我们做得很好，就这样不断迭代改进。

Yeah, I've been building companies for about 20 years now. I scaled a software company up pretty fast, sold it, scaled Archer up pretty fast, took it public. So every time I'm in this phase, I get to sit back and say, 'What did I learn from the past experiences and how do I do it better?' I figured we took a very differentiated approach to basically vertically design everything. I don't think there's any group in the world on the robotics side that designs more parts than we do on the robot. We design the motors, basically every part within there—the rotor, stator, everything. The sensors, the structure, the kinematics, the joints, the batteries that you saw today, the battery packs. That has really enabled us to control our destiny. We get to build our own supply chain. Without that, you're left at the mercy of some vendor, and if that has an issue, how are you going to solve it? If it's got a code problem, do you understand it? Can you QA it? Can you fix it? Can you patch it? So we understand the whole stack from top to bottom. It was an enormous lift up front to get the right people here that could do that. And then we've now been iterating through to the point where we have decently reliable systems now that run really well. I self-funded the whole company up front. We got to a million a month of burn in four months. It was no joke. We had a 40-person team in four or five months. They were very good. And then just here, 100 hours a week just trying to make it work with the team. We've made some mistakes. We've learned a lot. We've done some things well and just recursively getting better.

你为什么离开 Archer？

Why did you leave Archer?

机器人领域的元问题是解决人形机器人。如果你能解决这个问题，它将创造出世界上最大的商业，而且是远超其他。全球 GDP 的一半——略低于一半——来自人类劳动。我想去研究这个机器人的圣杯。在 Archer，我负责了我们所有飞机的设计。我觉得现在是这个十年，我们要把人形机器人带给大众。这可能是我们一生中最重要的商业之一。所以我现在能够从事我认为是我整个职业生涯中最重要的领域之一。在 Archer，我组建了整个团队，领导了所有飞机的工程设计，并带领公司完成了上市。我们现在处于一个很好的位置，可以在联邦空域认证飞机。而 Figure 也处于一个很好的位置，可以真正为世界扩展物理智能。

The meta problem in robotics is to be able to solve a humanoid robot. If you can solve this, it'll build the biggest business in the world by a large factor. Half the world's GDP—a little under half—is human labor. I wanted to go work on building this holy grail of robotics. At Archer, I led design for every aircraft we have there. I feel this is now the decade we're going to bring humanoid robots to the masses. It's probably one of the most important businesses of our lifetime. So I get to work now on what I think is one of the more important areas of my whole career. At Archer, I built the whole team up, led all engineering design for all the aircraft, and led the company through public offering. We're now in a good spot to certify the aircraft in federal airspace. And Figure is also in a good spot here to really scale up physical intelligence for the world.

我最近请了 a16z Perennial 的 Michelle Dell Buono 来节目。那是 Mark 和 Ben 的家族办公室。我们聊了流动性事件，以及作为创始人，你在第一次、第二次或第三次流动性事件时会怎么做。你为什么在 IPO 之后决定再创办一家公司？

I recently had Michelle Dell Buono from a16z Perennial on. It's Mark and Ben's multi-family office. We were talking through liquidity events and what you do as a founder for your first liquidity event or second or third. Why did you decide after the IPO to start another company?

嗯。

Yeah.

并且自己出资。

And fund it.

是的。实际上，从那以后我还创办了其他几家公司。简单来说，我关注人形机器人领域已经几十年了。我们很久以前就看到人形机器人了，但它们一直走在错误的道路上。我们造错了东西，或者只是业余水平，或者工程决策不正确。我只是觉得有必要更快地推进这个领域。我在 Harkin Cover 也在做同样的事情。我还有另外几家公司，我觉得情况类似——如果交给世界去做，我不确定我们会不会朝着正确的方向前进。对于人形机器人，四年前我们最好的成果就是波士顿动力的液压人形机器人 Atlas。它到处漏油，只能运行 20 分钟。它非常大，非常不安全。你永远不能把它放在人类旁边。那是经典的控制方法。而且，波士顿动力确实有深厚的研究传统，但不是商业化，或者说商业化程度不高。所以我只是觉得需要有一个团队进来，真正把这件事推向大众。如果没有我介入这个领域，我不知道我们是否能达到现在的水平。也许我们会达到，我们拭目以待。但我认为 Figure 已经真正证明了我们能够把时间线提前，进入现实世界。而且，就像我说的，我认为这是一项重要的业务。

Yeah. I started a few other companies actually since then as well. The short story is I've been watching the humanoid space for a couple decades. We've been seeing humanoids for a long time. They've just been on the wrong vector. We were building the wrong stuff, or we were doing it in a hobbyist grade, or the engineering decisions were not correct. I just felt there was a need to advance the space much more rapidly. I'm doing the same with Harkin Cover. I have a couple other companies where I feel it's a similar situation—if left to the world to go do it, I'm unclear if we would head in the right direction. For humanoids, the best we had 4 years ago was Boston Dynamics with a hydraulic humanoid called Atlas. It dripped oil everywhere, lasted like 20 minutes. It was very big, very unsafe. You could never put it next to humans. That was classical controls methodology. And it was like, man, you need to—Boston Dynamics is really a large heritage around research, not commercialization, or not as much commercialization. So I just felt there was a need for a group to come in to really send this thing to the masses. Without my intervention in the space, I don't know if we would have gotten there. Or maybe we will—we'll find out. But I think Figure has really shown that we've been able to push the timelines left now to get into the real world. And like I said, I think it's a significant business.

Sorcery 节目由 Brex 赞助，这是一个被超过 30,000 家公司信赖的金融技术栈，包括美国三分之一的风险投资支持的初创公司。近 40% 的初创公司因为资金耗尽而失败。Brex 就是为了帮助创始人避免这种情况而生的。与让您的资金闲置、用费用蚕食它的传统银行不同，Brex 旨在帮助您更明智地消费并更快地行动。他们的一体化解决方案将支票账户、资金管理和 FDIC 保护整合到一个强大的账户中。您可以闪电般快速地在全球范围内收发资金，通过他们的合作银行获得标准 FDIC 保险的 20 倍覆盖，甚至从第一天起就能获得高收益。凭借当日甚至当小时的流动性，您可以随时使用您的资金。Scale AI、DoorDash、ServiceTitan、Hims、Anthropic、Flexport、Robinhood 和 Plaid 等公司都信赖并使用 Brex。今天就访问 brex.com/sourcery 开始吧。网址是 b r e x.com/sourcery。Turing 正在用需要真实专业知识和现实世界判断力的任务来训练下一代 AI。这就是为什么 Nvidia、Anthropic、Salesforce 和 Gemini 等公司都与 Turing 合作。Turing 基于真实操作轨迹构建逼真的强化学习环境和数据系统，这是前沿实验室训练超级智能所需的基础设施。请访问 turing.com。

Sorcery is brought to you by Brex, the financial stack trusted by more than 30,000 companies including one in three venture-backed startups in the US. Nearly 40% of startups fail because they run out of cash. Brex is literally built to help founders avoid that. Unlike traditional banks that let your money sit idle, chipping away at it with fees, Brex is designed to help you spend smarter and move faster. Their all-in-one solution combines checking, treasury, and FDIC protection into one powerful account. You can send and receive money globally at lightning speeds, get 20 times the standard FDIC coverage through their partner banks, and even high yield from day one. With same-day and even same-hour liquidity, access your funds anytime. Companies like Scale AI, DoorDash, ServiceTitan, Hims, Anthropic, Flexport, Robinhood, and Plaid trust and use Brex. Start today at brex.com/sourcery. That's b r e x.com/sourcery. Turing is training the next generation of AI with tasks that require real expertise and real-world judgment. That's why companies like Nvidia, Anthropic, Salesforce, and Gemini partner with Turing. Turing builds realistic reinforcement learning environments and data systems based on real operational traces, the kind of infrastructure frontier labs need to train superintelligence. Visit turing.com.

你怎么还有时间同时做 Harkin Cover？

How do you have time to also work on Harkin Cover?

诀窍就是不睡觉。

The trick is just not sleep.

好吧。你是有什么秘诀，还是你就是不睡觉？

Okay. Do you have an eight sleeve or you just don't sleep?

就是不睡觉。

You just don't sleep.

另外，请解释一下这些公司是做什么的。

Also, explain what those companies do for people.

好的。Cover 基本上是在为 K-12 学校设计检测系统。过去 10 年，美国校园枪击案增加了 10 倍。就像真的有人开枪了，情况已经完全失控。我的看法是，这是一个感知问题。我们需要在学生进入学校时，看看他们身上有没有带枪。如果有，就把枪从学生身上拿下来。如果没有，就让学生进去。大约 10 年前，NASA 喷气推进实验室设计了一种技术，可以从 5 到 20 米外检测出衣服下、背包和袋子里的武器。他们为伊拉克和阿富汗战争建造了它，用于在安全距离外发现自杀式炸弹背心和其他爆炸物。可以把它想象成机场的 L3 扫描系统，但那种只能从几英尺外进行。如果能做到 10 倍远的距离，你就可以基本上在人们进入学校时扫描每个人。而且这不仅仅是学校的事。你可以在世界上每个公共场所使用它。这是我的热情项目。我一直在关注这个领域。这个极客话题叫做太赫兹成像雷达。这就是我们在 Cover 做的事情。我有一个团队，大部分来自 NASA 喷气推进实验室。我拥有 NASA 喷气推进实验室的知识产权。我两年前把它剥离出来，并且正在资助这个项目。

Okay. So, Cover is basically designing detection systems for K-12 schools. School shootings in the US have gone up 10x in the last 10 years. It's like a weapon's been fired. And it's just basically gotten fully out of hand. My view is that it's a perception problem. We need to see if students have guns on them when they're entering schools. If they do, get the guns off the students. If they don't, let those students go in. There's a technology that was designed at NASA Jet Propulsion Lab about a decade ago that can detect weapons underneath clothes and in backpacks and bags from 5 to 20 meters away. They built it for the Iraq and Afghanistan war to find bomb vests and other explosives from a standoff distance. Think of it like the L3 scanning systems at an airport, but you can only do that from a few feet away. If you can do that 10x further away, you could basically scan everybody as they're coming into schools. And it's not just a school thing. You could use it at every public venue in the world. It's been my passion project. I've been following this space. The nerdy topic here is called terahertz imaging radar. That's what we do here at Cover. I have a team, most of which are from NASA Jet Propulsion Lab. I own the IP from NASA Jet Propulsion Lab. I spun it out 2 years ago and I'm funding this project.

哦。

Oh.

是的。

Yeah.

你是怎么得到的？

How did you get it?

哦。他们会卖给你的。是的。所以他们把它卖给了我。

Oh. They'll sell it to you. Yeah. So they sold it to me.

100%？你拥有 100%？

100% you own it 100%?

是的。

Yeah.

好的。

Okay.

然后加州理工学院，基本上就是 NASA 喷气推进实验室，在这项业务中持有非常小的少数股权。而且当初做这个项目的很多核心团队成员现在都在 Cover 和我一起。太棒了。我们得到的图像，这是一个硬件和 AI 问题。这是一个必须用 AI 解决的视觉问题。我们现在已经有原型在运行了。我们希望年底前能部署到第一批学校进行测试。全美有 13 万所 K-12 学校。我们需要制造并交付大量的设备。我已经自筹资金资助那家公司两年了。我们正在取得惊人的进展。我在帕萨迪纳有一个团队，就在洛杉矶的喷气推进实验室旁边。他们负责资助那部分。然后我还有一家独立的公司叫 Hark。那是我大约 7-8 个月前创办的 AI 实验室。它试图设计高度个性化的智能。我们在设计下一代 AI 模型，也在设计下一代与 AI 交互的 AI 设备。现在我们通过 20 年历史的电脑与 AI 交互，比如手机和 MacBook。它们不是与 AI 交互的理想中介和界面。我们两周前刚结束隐身模式。我们有一个 50 人的团队。我们在构建真正令人惊叹的、神奇的 AI 模型，也在构建真正神奇的硬件。

And then Caltech, which is basically NASA JPL, has a very small minority interest in the business. And a lot of the core team that did that is on my team now at Cover. It's awesome. The images we get, it's a hardware and AI problem. It's a vision problem that you have to solve with AI. We have prototypes running now. We will hopefully deploy to our first schools in beta by end of year. There are 130,000 K-12 schools. There's a huge amount we have to manufacture and get out the door. I've been self-funding that company for 2 years. We're making incredible progress. I have a team in Pasadena, right next to Jet Propulsion Lab in LA. They're funding that. Then I have a separate company called Hark. It's an AI lab I started about 7-8 months ago. It's trying to design highly personalized intelligence. We're designing next generation AI models and also the next generation of AI devices to interact with AI. Right now we interact with AI through 20-year-old computers, like phones and MacBooks. They're not the ideal intermediary and interface to AI. We just came out of stealth 2 weeks ago. We have a team of 50. We're building really awesome, magical AI models and we're building really magical hardware.

你最近因为对 Figure 与 OpenAI 合作的评论而走红。发生了什么？

You recently went viral for comments on your OpenAI partnership with Figure. What happened there?

是的。OpenAI 几年前领投了我们的 B 轮融资。作为其中的一部分，我们签订了一项合作协议，共同开发下一代 AI 模型。这很棒。我和那边的团队关系很好。Sam 和其他人。他们领投了我们的融资。他们引入了 Satya 和微软。他们共同领投。然后我们花了大约一年时间合作，研究如何让 AI 模型在人形机器人上工作。或者如何让语言模型在人形机器人上工作。他们对机器人技术非常感兴趣，而我们非常感兴趣的是如何让语言模型在机器人上运行。它们扮演什么角色，或者它们在机器人技术中是否发挥作用？我们和他们合作了一年。都是很好的人。我几乎每天、每周都和他们一起工作。到了某个时候，我们内部设计这些模型的团队已经远远超过了 OpenAI。我们在这方面要好得多。我们在机器人上测试、训练模型，所有这些都更好。我的团队有超过十年的机器人学习背景。我想，当 OpenAI 看到我们进入机器人领域时，他们也有一些兴趣。所以我解雇了他。

Yeah. OpenAI led our Series B a couple years ago. As a part of that, we did a collaboration agreement to work on next generation AI models together. It's great. I got to know the team really well over there. Sam and the rest of the group. They led our round. They brought in Satya and Microsoft. They co-led it. Then we spent basically a year collaborating on how to get AI models to work on a humanoid. Or how to get language models to work on a humanoid. They were very interested in robotics, and we were very interested in understanding how to get language models on robots. What part do they play, or do they play a part in robotics? We spent a year working with them. Nice folks. I was working with them almost every day, every week. It got to a point where our team internally that was designing these models were running circles around OpenAI. We were just way better at this. We were better about testing on the robots, training the models, all of it. My team had come from robot learning backgrounds for over a decade. I think there was also some interest as OpenAI was watching us getting into robotics. So I fired him.

为什么一开始让他们投资？

Why did you let them invest in the first place?

我和 Sam 以及团队关系很好。我认为我们双方可能有很多潜在的战略利益，比如如何开发其中一些系统并相互学习。结果发现我在这方面有点错了。

I got to know Sam well and the team. I thought there could be a lot of potential strategic interests from both of us, like how to develop some of these systems and learn from each other. It turned out I was kind of wrong on that.

是那时候，还是甚至更早，你开始在这里变得更加注重安全？因为即使进来，我的手机也被遮住了，有受限区域，对知识产权非常严格。是什么……

Is that when, or even before that, when you started to become more secure here? Because even coming in, my phone is covered, there are restricted areas, it's very close on IP. What is...

是的。我们一直都很注重安全。我认为我们做的事情知识产权风险很高。我们非常仔细地考虑我们的工程 CAD 和软件，确保从网络安全和内部安全的角度来看都非常安全。我们的办公室非常开放。你可以从你的电脑上看到很多东西。

Yeah. We've always been pretty secure. I think what we're doing is very high IP risk. We really think carefully about our engineering CAD and software, making sure it's very secure from a cybersecurity perspective and internal security perspective. Our office is really open. You can see a lot of stuff from your computer.

是的，所以我们以前有人进来就随便拍照。你会说，“哇，这可不行。”比如，我不喜欢硬件之类的东西放在人面前。所以我们……

Yeah, so we used to have people come in and just snap photos randomly. You're like, "Whoa, that's not okay." Like, I don't like hardware or something in front of the person. So we like...

在湾区，有很多蜜罐、间谍。

In the Bay Area, there's a lot of honey pots, spies.

有一天我们抬头看，就在这个办公室里，我们看着窗户顶部的角落，有一架无人机停在那里，正往办公室里看，就在那边。

One day we're looking up, and we're in this office, and we look at the corner of the window at the top, there's like a drone sitting there looking in the office, right over here.

相信吧。

Believe it.

主要区域，是的。我们想，“天哪，我们得在这里改变一些东西了。”

Main area, yeah. We're like, "Oh my gosh, we got to change some things here."

你查出那是谁了吗？

Did you find out who that was?

我们没有查出是谁，但我们给所有玻璃都贴了膜。我们现在有非常严格的安全措施，无论是物理还是数字。我的意思是，我们在设计一些疯狂的东西，所以我们要不惜一切代价保护它。不过，我们一直是这样。

We didn't find out who it was, but we tinted all the glass. We have a really strict security, both physical and digital now. I mean, we're designing some crazy stuff, so we want to protect it at all costs here. But no, we've always been like this.

有点偏执，这总是有帮助的。

A little paranoid, which always helps.

典型的创始人特质，对吧？

Typical founder trait, right?

是的，完全正确。

Yeah, exactly.

说到这个，我们的赞助商之一是 Brex，他们关注绩效、更明智的支出、更快的行动。我很好奇，作为领导者，你如何保持自己的绩效，无论是心理上还是团队领导方面？

Speaking of that, so one of our sponsors is Brex and they're about performance, spending smarter, moving faster. I'm curious from your standpoint as a leader, how do you maintain your performance whether it's like mental or like team leadership-wise?

是的，你之前问过我怎么能有时间做所有这些事。你知道，我觉得我的生活里一直有三个时间桶。我觉得我有家庭，有工作，还有和朋友以及认识的人一起做的事。比如和朋友年度旅行，或者打高尔夫，或者大学里的其他事。大约在 Archer 的时候，我到了那种“天哪，我真的没有时间再做这三件事了”的地步。所以，大约 5 年前我决定不再参加年度高尔夫旅行。也不再因为某个十年没见的人来了就要出去吃饭。我不再那样做了。我的时间要么花在家庭上，要么花在我的公司上。这就是我做的全部。这其实挺好的。所以，这就像是我真正关心的事情，我可以全身心投入，把它做好。所以，在这里，就像你一样，我昨晚在办公室待到午夜，但我每晚都回家和孩子们吃晚饭。所以我 6 点前到家，吃晚饭，哄他们睡觉，如果需要的话再回来，或者待在家里，看情况而定，但通常我会回去工作。并且确保我解决这些公司里最棘手的问题，帮助它们 Scaling。所以，不管怎样，从时间的角度来看，是的，我确实必须非常谨慎地考虑我做什么。即使现在在办公室，我也必须非常谨慎地考虑我把时间花在什么上。所以，随着时间的推移我学到的是，你来到这里时问过我，你有没有角落办公室，还是你在……

Yeah, you asked before like how do I have time to do all this stuff? You know, I think I always had these three buckets of time in my life. I feel like I have my family, I have work, and I have things you do with friends and people you know. Whatever, annual trip with friends or golf or whatever else it is from college. About, you know, when I was at Archer, I got to the point where I was like man, I don't really have time to do all three anymore. So, I decided like 5 years ago to stop doing the annual golf trip. And stop doing like this person's in town I haven't seen in 10 years, we need to go out to dinner. I just don't do that anymore. It's been all my time with family or on my companies. That's all I do. And it's kind of nice. So, it's like the stuff I really care about and I get to kind of go all in on it and do a really good job. So, here I'm like you I was at the office last night till like midnight, but I come home every night to have dinner with my kids. So, I'm at home by 6:00 and I do dinner and do bedtime and then come back in if I need to or stay home depending on what's going on, but usually it's back to work. And just make sure I unblock the most pernicious problems at these companies and help scale. So, anyway, I think from a time perspective, yeah, I kind of have to be very thoughtful about what I do. When I'm even in the office now, I got to be really thoughtful about how I spend my time on. So, what I've learned over the time, you asked when you came here is like, do you have the corner office or do you have a you're in the...

牛棚（开放办公区）。

The bullpen.

牛棚。天哪，我有一个……所以，你知道，在我们让 Archer 上市后的一天，我醒来时想，我感觉自己像在《土拨鼠之日》电影里，我和所有高管每天都在那个会议室里。而且，你知道，我通常在产品工程一线帮助设计飞机，但我却被困住了。我被困在每季度两天的董事会会议、分析师电话会议，和我的首席人力资源官或总法律顾问讨论各种事情，我心想，这里肯定有什么不对劲。我的意思是，就像这里有一个领导办公室，俯瞰着所有人，而不是在地面上做真正的工作。而且，你知道，大约在那个时候我做了个决定，基本上把我日程上的所有事情都清掉，只把时间花在产品工程上。所以，我做几件事，比如你是第一个真正看到整个办公室像这样（开放）的人。

The bullpen. Man, I have like a... So, you know, one day after we took the car, Archer public, I woke up one day and I was like, I feel like I was in this Groundhog movie where I was in this conference room with all my C-suite and was in there like every day. And, you know, usually I'm like on the floor of product engineering helping to build aircraft design and I was just like stuck. I was stuck in like two day a quarter board meetings and analyst callbacks and talking about, you know, my CHRO or GC and all these different things and I was like, this is just something's wrong here. I mean, it's like a lead office over here like looking down on everybody, like not doing real work on the ground. And, you know, I made a decision kind of around that time to like just basically remove everything on my plate just spend time on like product engineering. So, I do a few things like you're the first person that's actually seen the whole office like this.

哇。

Wow.

这很酷。我们不常做这些，而且我喜欢你的节目，所以很高兴你能来，让世界先睹为快，看看我们在做什么。

Which is cool. We don't do a lot of these and I love your show, so it's like great to have you here and kind of give a sneak peek to the world on like what we're doing.

真不错。

Cute.

是的。然后，我剩下的时间都花在如何做产品工程上？如何推进人形机器人做更好的事情？或者如何把我们的系统推向世界？我们需要开始测试并获取反馈。以及如何把我们的模型和设备推向世界以实现 Scaling？我觉得这些才是真正重要的事情。传统公关和活动，比如参加贸易展、坐在小组讨论上之类的，这些都不重要。

Yeah. And, um, and then the rest of my time's on like how do I work on product engineering? How do I advance the humanoid to do better things? Or how do I make covers system out to the world? We need to start testing it and getting feedback. And how do we get heart models and devices out to the world to scale? Like those are the things I think really matter. Matters less about like doing traditional PR and event, you know, like going to like trade shows and, you know, sitting on like these like panels and stuff. None of this matters.

是的。

Yeah.

那不是真实的。所以，我尽量把时间花在牛棚里，然后我尽量最大化我在办公室内外的、花在重要事情上的时间。

It's not real. So, I try to spend my time in the bullpen and then I try to maximize my time both when I'm in or out of the office on things that are important.

Fundrise 旗下的 VCX，是私人科技公司的公开股票代码，让各种规模的投资者都能投资风险投资。在 getvcx.com 查看投资组合。网址是 getvcx.com。有些人可能还没听说过，我们的赞助商 Public 刚刚推出了一个叫“生成资产”的功能，它把 AI 带入了投资领域，说实话我从未见过这样的方式。它的工作原理是这样的：你输入一个想法，比如“具有正自由现金流的 AI 驱动的供应链公司”或“年营收增长超过 25% 的国防科技公司”。然后 Public 的 AI 会派遣一群智能体，扫描每一只美国股票，评估它们，并立即围绕你的论点构建一个自定义指数。最突出的是它清晰解释为什么包含每只股票。在你投资之前，你甚至可以根据标普 500 指数对你的想法进行回测，这样你就能在真实背景下做决策，而不是猜测。除了生成资产，Public 还让你在一个地方投资股票、债券、期权、加密货币。当你从其他平台转入投资时，他们甚至会给你 1% 的无上限匹配。如果你想建立一个真正反映你论点的投资组合，请访问 public.com/sourcing。

VCX by Fundrise, the public ticker for private tech, allowing investors of all sizes to invest in venture capital. View the portfolio at getvcx.com. That's getvcx.com. Some of you may not have heard this yet, but our sponsor Public just launched something called generated assets and it brings AI into investing in a way I've honestly never seen before. Here's how it works. You type in an idea like AI-powered supply chain companies with positive free cash flow or defense tech companies growing revenue over 25% year-over-year. Public's AI then dispatches a swarm of agents that scan every single US stock, evaluates them, and instantly builds a custom index around your thesis. What really stands out is how clearly it explains why each stock is included. And before you invest, you can even backtest your idea against the S&P 500, so you're making decisions with real context, not just guessing. And beyond generated assets, Public lets you invest in stocks, bonds, options, crypto, all in one place. They'll even give you an uncapped 1% match when you transfer your investments over from another platform. If you want to build a portfolio that actually reflects your thesis, visit public.com/sourcing.

由 Public Investing 付费。完整披露信息见描述。企业 AI 运行在 Merge 上，这是一个用于集成、智能体工具和模型编排的 AI 基础设施平台。所以，你的团队应该交付产品，而不是管道。Mistral、Dropbox 和 Drata 已经在生产环境中信任 Merge。在 merge.dev 开始构建。创始人在 Deel 上更快地 Scaling。在几分钟内为任何国家设置工资单，雇佣任何地方的任何人，快速处理签证，然后回到构建。访问 deel.com/sourcing。网址是 d e e l.com/sourcing。

Paid for by Public Investing. Full disclosures in the description. Enterprise AI runs on Merge, the AI infra platform for integrations, agent tooling, and model orchestration. So, your team should ship product, not plumbing. Mistral, Dropbox, and Drata already trust Merge in production. Start building at merge.dev. Founders scale faster on Deel. Set up payroll for any country in minutes, hire anyone anywhere, get visas handled fast, and get back to building. Visit deel.com/sourcing. That's d e e l.com/sourcing.

你提到了一些来自 Shopify 的灵感，但从领导力的角度来看，你钦佩或敬仰哪些人，他们帮助你前进并保持动力？

You mentioned a bit of inspiration on ideas around coming from Shopify, but on the leadership standpoint, like who are people that you admire or have looked up to that like help carry you forward and keep you motivated?

是的，我的意思是，有点像比史蒂夫·乔布斯和那一代人晚一代，比如杰夫·贝索斯，他是我们 Figure 的大投资者，你知道，我基本上可以接触到并与他交谈。而且，你知道，对我来说，我想以 11 分（满分 10 分）的水平来玩这个游戏。我想非常努力地去做这件事。所以我们非常认真。我是一个相当有竞争力的人。所以如果我要做这件事，并且牺牲我的生活和时间来投入，我就想把它做好。我想赢，我想把每件事都做到最硬核的水平。所以我认为我敬仰的人是那些做到了这一点、取得了成功，但真正把时间投入到成为那种情况下最佳运动员的人。所以，是的，我在这里看着自己。

Yeah, I was I mean, kind of like a generation behind like Steve Jobs and that whole, you know, crew, like Jeff Bezos, which is a big investor for us at Figure, and you know, I get to basically have access to and talk with. And you know, for me, like I want to play the game like 11 out of 10. I want to go really hard at this. Like so we're really serious. I'm pretty competitive person. So if I'm going to do this and like sacrifice my life to do this and my time, I just want to nail it. I want to win and I want to do everything to the most hardcore level possible. And so I think the folks I looked up to in the world are the folks that have done that, have been successful, but have like really devoted their time to being kind of best athlete in those situations. So yeah, and I'm looking at myself here.

比如说，我怎么让这些公司运转起来？没有什么比 10 年、15 年、20 年后这些事还没成更糟糕的了。我错过了所有的高尔夫之旅，错过了和家人在一起的时光，错过了生活中的所有这些。现在我已经有足够的钱了，所以我做这件事是因为我热爱它。而且我最好能不断进步，递归式地提升自己。我敬佩像史蒂夫和杰夫那样的人，他们比我大一辈，在很多领域都做得很好。他们也为这件事奉献了一生，一直是我的灵感来源。这很难，他们坚持了几十年。这种事不是一夜之间发生的。我做这行已经 20 年了，却觉得自己才刚刚开始。

Like, how do I make these companies work? There's nothing worse than 10, 15, 20 years from now these things not working. I missed all my golf trips, time with family, all this stuff in my life. At this point I have plenty of money, so I'm doing this because I love it. And I better be getting better at it, recursively improving. I look up to folks like Steve and Jeff, a generation above me, who've been good at many things across many areas. They've devoted their life to this and have been a source of inspiration. It's been hard, and they've done it over decades. This stuff doesn't happen overnight. I've been doing this for 20 years now, and I feel like I'm just starting.

这项业务最大的风险是什么？

What are the biggest risks to the business?

人形机器人这件事太难了，我甚至没法很好地解释。让机器人做到我们今天给你展示的那些事，几乎要了我的命。而且我们还有一场硬仗要打。Archer 也一样。我们必须每天在城市上空飞行，并且要确保它非常安全，成为最安全的交通方式之一。所以，我正在攻克这些事情，如果你看看这些事情成功的概率，那是非常低的。我觉得这大概就是事实。所以，我基本上有一个漏斗，里面装满了我每天必须解决的最棘手、最顽固的问题，它们真的非常难。我最大的风险——有一长串风险清单，都是可能伤害我们、让我们失败的事情。最重要的是能够在长时间内完成端到端的有用工作。我想把一个机器人放进一个家庭，让它成功完成 7 到 10 小时的工作，不出故障，无需人工干预，然后每天都这样，永远如此。这是个难题，从来没有人做到过。机器人非常复杂——就像从零开始设计涡轮风扇发动机、火箭或飞机一样。我们还从零开始设计了整个供应链。所以它有很多问题。第一次不会成功。我们正在解决这些问题。我们以前在这里运行 Figure One，那个机器人能跑大约一个小时然后就会出故障。我们知道它为什么会摔倒或断电的所有原因。我们转向了 Figure Two，我想我们大概每天看到一次故障。现在有了 Figure Three，它们都在这里整天运行。我们每周都会看到故障——不是每台机器人都有，但确实会有。我们有一个清单，正在逐步解决。但这真的很难。随着我们扩大机队规模，故障的绝对数量在上升，我们必须想办法解决这些。因为，正如你所说，机器人可能处于的状态数量太高了。你无法合理预测机器人在世界每个地方、每个时间点会是什么样子。对于汽车，你大概可以——你可以在路上行驶。但这也是个难题。总之，我想说的是，我们有一堆问题——对公司来说，这是一个永无止境的问题之城。我们必须以前所未有的速度进行制造。我们必须让人形机器人在没有人工干预的情况下自主工作。从来没有人做到过。我们必须让它与 AI 政策配合。硬件不能出故障。它必须非常实惠。我们必须制造很多。而且我们必须让消费者想要它们。

The humanoid thing is just so hard. I can't even explain it very well. Getting the robots to do the things we showed you today has almost killed me. And we have such an uphill battle to go do that. Same with Archer. We have to fly aircraft every day above cities and make it really safe, one of the safest forms of transportation. So this stuff I'm working through, if you look at the odds of these things working, they're super low. And I think that's probably pretty accurate. So I basically have a funnel of the hardest, most pernicious problems I have to solve every day, and they're really tough. My biggest risk—there's a long list of risks, things that could hurt us, make us not make it. The most important thing is to be able to do end-to-end useful work over a long time horizon. I want to put a robot into a home and have it do 7 to 10 hours of work successfully without failures, with no human intervention, and then do that every day forever. It's a hard problem. Nobody's ever shown that. The robot is very complicated—like designing a turbofan, a rocket, or an aircraft from scratch. We've also designed the entire supply chain from scratch. So it's got a ton of problems. It doesn't work the first time. We're working those problems out. We used to run Figure One here, and that robot could run for about an hour before it would fault. We know all the reasons why it would fall or lose power. We moved to Figure Two, and I think we saw that maybe once a day. Now with Figure Three, they're all running all day. We see faults every week—not on every robot, but we see faults. We have a list and we're working those down. But it's really hard. As we grow the fleet, the absolute number of faults is rising, and we have to figure out how to solve those. Because, as you said, the number of states the robot can be in is so high. You can't reasonably predict what the robot will look like at every timestamp everywhere in the world. You kind of can for a car—you can just drive on roads. It's a hard problem too. Anyway, what I'm trying to say is we have a funhouse of problems—it's never-ending problem city for the company. We have to manufacture at unprecedented rates. We have to get humanoid robots to work autonomously without human intervention. Nobody's ever shown that. We have to make it work with AI policies. The hardware can't fail. It's got to be really affordable. We've got to make a lot of them. And we've got to get consumers to want them.

这确实很多。而且你已经筹集了近 20 亿美元。最近公开的估值是 390 亿美元。你认为资本是风险或约束，还是估值是风险？

That's a lot. And you've raised nearly $2 billion. I think most recently publicly stated at $39 billion valuation. Do you see capital as a risk or a constraint, or the valuation as a risk?

这将打造世界上最大的企业。商业市场中，略低于 GDP 的一半是人力劳动——他们为人类支付工资。我们做的是人类的工作。所以，如果机器人运行良好，我们将有能力向商业劳动力市场投放数十亿台机器人。每年有大约 30 万亿、40 万亿美元的工资支付给从事这些工作的人。我们将能够扩展这些工作，自动化其中很大一部分，并继续扩大规模。再加上我们在家庭中看到的那些，将创造巨大的收入——数十万亿美元。你将打造一个庞大的东西。我的意思是，大多数公司的估值是多少？科技公司通常是收入的 10 到 20 倍。你面对的是 1000 亿到 1 万亿美元的收入。这将是一个巨大的业务。

This will build the biggest business in the world. Like, a little under half of GDP is human labor in the commercial market—they pay wages for humans. We do human work. So, if the robots work well, we will have the ability to ship billions of robots into the commercial workforce. There's like 30 trillion, 40 trillion of wages paid every year to folks doing that work. We'll be able to expand that work, automate a lot of it, and continue to scale it up. That, plus the stuff we're seeing in the home, will build enormous revenue—tens of trillions. You'll build something massive. I mean, what do most companies trade at? Tech companies trade at 10 or 20 times revenue. You're looking at 100 billion to a trillion dollars of revenue. This is going to be a huge business.

你有登月或去火星的计划吗？

Do you have plans for the moon or Mars?

我们很乐意把机器人送入太空。是的，让我们把它们送出去。

We'd love to send robots into space. Yeah. Let's get them out there.

好的。

Okay.

我有一份礼物要送给你。

I have a gift for you.

真的吗？

You do?

是的。我有一些 Figure 的周边。所以，嗯，一些帽子和 T 恤。你可以在这里代表 Figure 品牌。

I do. I have some Figure swag. So, yeah, some hats, shirts. So you can rep the Figure brand here.

这周边真不少。

That's a lot of merch.

很多周边。

A lot of merch.

哇，我得买个新衣柜了。这太棒了。有趣的是，我觉得他们在录制参观时没注意到，但我问过你标志是怎么做的，你说是机器人的脚步。

Wow, I'm going to have to get a new closet. This is great. It was funny, I don't think they caught this when we were recording the tour, but I asked you how you made the logo and you said it was the robot steps.

是的，基本上机器人就是这样走路的。即使在模拟中，它们看起来也像小方块。所以你有一个小小的行走轨迹，然后还有一个抽象的 F。

Yeah, it basically the robot takes footsteps like this. Even in simulation they look like little squares. So you have a little walk, and then you also have a little abstract F.

不太像。

It's not really.

好吧，我们保留行走的部分。

Okay, we'll keep the walk.

好的。那么，在我们结束之前，你对明年最期待的是什么？

Okay. Well, as we wrap up, what are you most looking forward to for this next year?

今年，我想把机器人规模化地推向世界。第二件我想解决的事情是我所说的通用机器人——一个能做人类能做的一切事情的机器人。你几乎会有一种像是穿着紧身衣的人的感觉。

For this year, I want to ship robots at scale out to the world. And the second thing I want to solve is what I call general robotics—a robot that can do everything a human can. And you have almost like a feeling of a human in a body suit.

你可以跟它对话，它能看着你，进行推理和视觉理解。你可以把它放到任何地方，它就能环顾四周、推理并理解。我们想解决这个问题。这对我们来说是个巨大的挑战。我们非常专注于让机器人走出实验室，解决通用机器人技术。我希望这里能成为我们第一个看到 AGI 出现在物理世界的地方。我们认为我们拥有实现这一目标的配方和正确的训练流程。今年和明年将是关键，看看我们能否攻克这一难关。

You can talk to it, it can look at you, reason, and understand visually. You can drop it into any place and it can just look around, reason, and understand. We want to solve that problem. That's a huge problem for us. We have a huge focus on getting robots out the door and solving general robotics. I want this to be the first place where we see AGI in the physical world. We think we have the recipe and the right training processes in place to do this. This year and next will be important to see if we can crack it.

太令人兴奋了。非常感谢你这一整天的陪伴，带我们参观了整个园区，还有这么深入的讨论。我真的很感激。

Exciting. Well, thank you so much for the full day here, the entire tour of the whole campus and the thoughtful discussion. I really appreciate it.

很高兴你能来。

It was great to have you.

嘿，我是 Molly。如果你喜欢我们的访谈，可以看看我们的通讯 sorcery.vc，我们每周发送一次顶级交易和科技头条邮件，并且会深入探讨我们的播客访谈。今天就订阅 sorcery 吧，别忘了在 YouTube、Spotify、Apple 或其他你收听的地方订阅播客。链接在描述中，点击即可注册。

Hey, it's Molly. If you enjoy our interviews, check out our newsletter sorcery.vc where we deliver a once-a-week top deals and tech headlines email and also go deeper on our podcast interviews. Subscribe to sorcery today and don't forget to subscribe to the podcast on YouTube, Spotify, Apple, or wherever you listen. Link in description to sign up.