iPhone包装揭示的机器人市场机会

No matter what anybody says, today in 2026, we have a highly interconnected, interdependent global supply chain. Both U.S. companies and Chinese companies are thinking very hard about what the long-term strategy looks like, and ultimately, everybody is trying to build their industrial capacity.

We see a lot of robot pitches. Very few people are thinking about what the robot is supposed to do at the end of all of this.

Almost everybody we’re selling a robot to has never purchased a robot before. Almost all of the market is greenfield. The reality is, you need to build a full-stack platform. Tutor robots have helped manufacture and ship over 250 million products to American consumers. We’re training frontier robot AI models, manufacturing robots at scale, and directly deploying them into the largest businesses in the supply chain.

How does product innovation work in robotics? Most companies are just on V1, whereas you’ve got 2 versions of the robot and you continue to iterate really fast.

Josh—

Yes.

Thanks for joining us. We’re with Josh Gruenstein, who is the CEO and founder—

Co-founder.

—of Tutor Intelligence, and we’re going to be talking about robots today.

Mm-hmm.

Before we jump in, maybe give us a quick background on yourself and the company.

1. The Full Stack Robotics Thesis

Yeah. I’m Josh. I started Tutor about 5.5 years ago with the core thesis that if you could start deploying robots out into the world, into all the places in the physical economy where robots are needed most, you could start to build the datasets, the models, and the supply chain needed to proliferate robots throughout the physical economy.

We’re about 160 people, based primarily in Boston, and we’re training frontier robot AI models, manufacturing robots at scale, and directly deploying them into the largest businesses in the supply chain.

Talk a little bit about this: we see a lot of pitches. There are a lot of companies that talk about doing robotics but don’t actually deploy robots, or haven’t yet.

Yeah.

You’re one of the few companies that’s actually deployed robotics at scale. Maybe talk about the state of the business.

We have robots deployed across the United States, everywhere from small mom-and-pop businesses to Fortune 50 and Fortune 100 businesses. Claude told me last night that it estimates Tutor robots have helped manufacture and ship over 250 million products to American consumers.

So presumably, most American consumers are Tutor customers a couple of layers down. I think this comes from the origin story of the business, which was us looking at the state of robot learning. It was obvious to us that robot intelligence and robot software were the bottleneck. For 60 years, robotics has been about custom-engineered machines that can’t scale.

Right.

We started the company with the thesis that if we could start deploying robots, we could build the labeled datasets that help us train these models and deliver this value to the American consumer.

We see a lot of robot pitches. 99% of them start with some insight about the model and how you’re going to build the model, or some insight about the data—the shape of the data, that is. Very few people are thinking about what the robot is supposed to do at the end of all of this. You started from the reverse, and you had the insight that you could actually tutor the robot, so to speak.

Yeah.

Hence the name of your company. Where did that insight come from?

It came from this thought experiment. We had this insight that there are 8 billion people on Earth who all have this baseline physical intelligence that we want our robots to have. Imagine you had infinitely powerful software that could connect any human being on Earth to any robot and embody that robot with intelligence. That’s like this deus ex machina where suddenly you have unlimited, infinite robot intelligence. Let’s say you take that for granted. Now what? The problem isn’t actually solved. You still need to manufacture robots at scale. Robots are a cottage industry today in the United States.

Robots are a rounding error even in the most automated industries, like manufacturing and logistics. That doesn’t exist. Certainly, none of the software exists, and none of the customers exist because, again, the robots can’t do any of the jobs that customers want robots to do. Robotics isn’t any one layer of the stack. It’s not just about intelligence, hardware, or the customer. It’s about figuring out how to put all 3 pieces together, and it’s the combination of all 3 that lets you build, in my view, a successful platform at any one of those layers.

Yeah. Internally, we talk about robotics being the ultimate systems company—

Yep.

Yeah.

—as opposed to being one or the other.

What’s funny about this, though, and it’s a question I wanted to ask you: you’ve got a pretty deep technical background, right? MIT robotics—is that fair? But whenever I talk to you, I feel like I’m talking to an operations guy who talks about unit economics, operations, and supply chain.

And even a go-to-market guy.

And even a go-to-market guy. So it’s like the transformation is complete. Maybe this is an open-ended question, but I’ve always wondered: is the harder problem here actually that stuff?

No, but I think you have to do all of it. When I was just a roboticist—

Yeah.

—and only working on the technology side of this problem, I think my strength as a roboticist was that I was very full stack, and I was willing to think about the manufacturing and the hardware. I was willing to think about FPGAs and ASICs. If you view robotics the same way I do—as this ultimate systems problem—the only way you can make forward progress is by figuring out all the clever trade-offs across as many layers of the stack as you can.

If you throw out one of the layers of the stack and say, “Well, we’re not going to touch that,” you’ve eliminated 90% of your trade-off space to make progress.

Oh, interesting.

Starting a business, it became really clear to me that robotics isn’t just full stack in terms of model, hardware, and software. It’s full stack in terms of model, hardware, software, supply chain, customers, and product—

Go-to-market.

—right? It’s like—

Cash. Yeah.

Exactly. And similarly—

Business model.

—you ask the question, “Why hasn’t robotics worked?” There have been a lot of robot startups. I would argue that no robot startup has been really massively economically successful. Why is that? I think it’s because, first, you don’t have this core primitive, which is intelligence, and that lets you bring the unit economics of the system to a certain level. Second, it’s a super-disaggregated market.

You have infinite players, and they’re all trying to work with each other. The reality is that what we need in robotics is to be able to make trade-offs across the full stack.

It’s interesting. This is an observation we’ve had, and it’s interesting to articulate it this way: most robot companies start horizontal, and they all go vertical. Every one of them. They literally start with, “We have a model,” or whatever, and then they start buying 3D printers.

Do you think that’s because the market is so immature, like what you had to do in the early days of cars? Or do you think it’s inherent in this domain? Even in 20 years, will you have to verticalize because the economy of movement or something requires that?

No, I think it’s a maturity in the market, and you can see similar things in the early computing market. To us, the question then becomes: how do you build general platforms that let people build businesses that aren’t robotics businesses? We have all these incredible software businesses today, and the people who run them don’t have to think about chip architectures. How do we make that possible in the physical world with robotics, and how do you go about building that platform?

I agree with you. People have tried to solve that problem by thinking about very narrow layers of the stack. In our view, the reality is that you need to build a full-stack platform, and the only way to know that full-stack platform really works is to have successful business units operating on top of it.

Are you open to the possibility that the physical world is different from the digital world, and that the reason we horizontalized compute is that you have clean abstractions that are generally applied—

Yeah.

And that’s just not the case for navigating physical space?

I think the beauty of something like a humanoid is that it’s the same clean abstraction, right? You think about what a CPU is: it’s a general computing machine. It’s not perfectly optimized for every use case, but it’s general enough that you can run most computing workloads on it.

Oh, but that’s your horizontal component within the humanoid.

Exactly.

But, for example, would you use the same brain or subcomponents to make a humanoid as you would to make a drone or some other form factor?

No, but you can make that same analogy in compute, right?

Got it. Got it.

We have CPUs, we have GPUs, we have ASICs, right? So I think you track the maturity of the market, and you start in this world of specialization and custom machines, because that’s the only thing the market can support.

Yep.

You move towards general purpose, and in robotics, this is going to look like humanoids and similar form factors. Then eventually you kind of break past to respecialization as the market is big enough.

Why is this thing called Tutor Intelligence?

Tutor was founded with this thesis that there is a whole internet of human intelligence that is networked, right? It’s posting on social media, it’s writing emails, and that network of human intelligence for LLMs and other kinds of digital world models has, by natural omission, created this beautiful dataset that we can use to train general-purpose models.

The issue is our work in the physical world is not similarly networked, right? I pick up my coffee cup, and nobody knows. Nobody recorded all the pressures of my fingertips, where they were, or the poses. So we don’t have this asset.

The core idea behind Tutor is: What if, instead of trying to find an algorithm where we could build robot intelligence from no data, we built this network of robots out in the real world? For us, trying to build intelligence from no data seems like a problem that’s rigged against us. That’s not how this worked in computer vision. It’s not how this worked in NLP.

We could connect the robots with human teleoperators, which we call tutors, who teach the robots how to do new tasks. In the same way the internet, by omission, created this dataset to train these frontier models, we can do the same with robots and build this internet of robot training data.

There are all these beautiful emergent properties. Suddenly, it’s not just a massive capital project. It is a project to build this network, but you’re also creating incredible economic value as you do it. You’re making everybody’s goods cheaper as you deploy these robots and collect this data. So it’s very positive-sum in this unique way.

When we thought about what to name this business, it’s not strictly artificial intelligence. There’s some artificial intelligence, but ultimately, it was this tutor intelligence that we were building, connecting and controlling these robots.

That’s an amazing insight. The funny thing is, when we talked to your customers before we invested, most of them didn’t seem to know or even care that there was a tutor, or when the tutor was operating and when your learning model was operating. Credit to you guys that you’ve made those very seamless, too.

To me, this is a signal that the robots are good, right? You want the feedback to be, “My shampoo that I’m manufacturing at volume is now 5 cents cheaper than it was.” If the feedback is, “The robot is pretty,” it’s like, “Oh, this is probably a really bad robot.”

Is the goal to not have tutors in the future, or does it matter?

We think about it at a very pragmatic level: What is the operational complexity of tutors? What is the economic impact of tutors? Think about the cost to serve models with an H200, and you can map that out. What’s the duty cycle and utilization of that H200? You can think very similarly about human intelligence: What is the heterogeneous profile of intelligence that you really want from humans that you can’t yet get from your models?

Then it becomes a unit economics question. How much engineering do you have to do to get that last nine out of your models versus using tutors? Going back to what we were talking about, you want to be able to make as many trade-offs here as you possibly can to make a business like this work.

I mean, it’s tempting to say that the economic marvel that happened with generative AI can also happen here. The marvel that happened with generative AI is that you train these models with some distribution of data, and it turns out basically the same distribution is economically viable.

Yeah.

Right? You train it on images, and people want images.

Yeah.

You train it on text, and people like it. That kind of distribution in, distribution out. It’s not clear that this is actually the case with these robots, right? You could train it on a distribution of data, but what you actually need might be just slightly outside that distribution. So I don’t know. You probably have the best visibility into how this scales, what the convergence is, and whether you can actually remove tutors. What’s your intuition on that?

I think this is one of the big questions, and I think it ultimately comes down to what the data is and where you’re getting it from.

The magic of the internet is that if you’re a company and you provide, at the application layer, tools for software engineers to accelerate their software development, you are naturally in the flow of all this really economically valuable work. If you can make money as a business, then you’re probably also making data as a business. Those economic incentives are aligned.

Similarly for robots, if you’re just going to strap cameras on everybody’s heads and send them throughout Indonesia to collect data doing X, Y, and Z, you have no guarantee.

Yeah.

You can guess, but there’s no invisible hand of the market guiding you to ensure that that data is actually economically valuable.

Right.

But if it’s an emission byproduct—if you have robots in real customer sites that customers are paying for to do that task—you know for sure that that data is economically valuable.

Doesn’t it kind of depend on the dimensionality of the solution set, though? It could be the case that the economically valuable stuff is so high-dimensional that you could solve 90% of it, but the last 10% you’ll need a human being to do for a while, just because getting that dataset is going to require too many tutors over too long a period of time.

Yeah, and that might be fine, right? You might be willing to accept that on the economics. That also might be a place where other forms of data matter. I think we’re not purists. We’re willing to take any trade-off, right? We love using simulation to widen these data distributions. We love using internet-scale data to widen these distributions. So we have a lot of tools in our toolbox other than real-world robot data.

And talk a little bit more about the simulations and these other tools, and the impact models like Astra are having on that. That seems to be another cottage industry of excitement for the robotics world, right? How are you using that?

It’s a sea change, right? I think we lived in this world for a while where everybody knew that all this general knowledge about the world was baked into these models, but they couldn’t really express it.

They didn’t have that last mile of spatial reasoning to connect the dots from what a GPT model knows to what an action diffusion transformer head can know. That’s changed. The labs are now post-training their models on 3D spatial data, like CAD and 3D modeling, which they’re doing because there’s a direct economic incentive and a real market for it.

So they can immediately monetize it, and it turns out that’s the exact spatial reasoning our models need. I think we’ve been catapulted many years ahead in the future, because we’re starting to be able to use all of that general intelligence that lives in these big frontier models for robots.

And because it’s a systems problem, then that effect for you is accelerator-y—

Yeah.

As opposed to slowing you down or saying somebody’s going to come and eat your lunch.

Yeah. No, we’ve seen a couple of these happen, right? We started this business in 2021. We were initially doing this with convolutional neural networks. We’ve gone through every sea change, and the thing that’s always valuable is the flow of data.

Yeah.

Right? The flow of data is always the thing that you can uniquely do. The million architectural tricks are worth doing if they help you accelerate the flywheel—if they help you deploy more robots, collect more data, manufacture at a bigger scale, or do more tasks. But ultimately, we think the thing with long-standing generational value in a business is the flywheel and the loops themselves.

Can you talk a little bit about what you mean when you say you build robots?

Yeah.

At one end of the continuum, you buy a Unitree.

Yes.

You prompt it. On the other end, you’re machining parts. Where are you in this?

For context, in 2025, the United States deployed 38,000 industrial robot work cells, which is tiny compared to roughly a million—

You’re not counting the FANUCs and the old-world robots.

The FANUCs, right. But that’s it. Nobody deployed any Unitrees in the United States. They would have shown up in those numbers, but there were probably zero.

The supply chain that exists for robots is both too small and, as you called out, wrong. The robots that FANUC and ABB are building are mostly not made in the United States. But even when they are, they’re building objects designed for systems integrators and fast, repetitive motions. They’re not cut from the set of trade-offs you’d want to co-design with frontier models.

So we’re starting from scratch on the hardware side. In order to deploy robots at scale, Tutor has actually been forced to design and manufacture our own hardware. Today, we’re not building actuators or cameras. There’s a whole layer of stuff we source from OEMs that manufacture at a much larger scale. But we build the actual robot—the thing that drives around, has the arm on it, and can do all the tasks—at scale in Watertown, Massachusetts.

Can you talk about the morphology of the robot? Is that called morphology? Anatomy?

You can use whatever word you want.

It’s called anatomy.

It’s—

What do they look like?

What do they look like? We make 2. One robot we make—

Look at his website, Machin. Yeah, go ahead.

Yeah.

Sorry.

So, Cassie. Cassie is kind of a brute, designed to do heavy-duty work. It has a single arm and a lot of suction cups, and it’s powered by foundation models in the 200 million to 400 million parameter range. It came out of the work Tutor started in 2021: What can foundation models do for robots? They can predict affordances and grasps that you can then execute.

It’s a co-design of the hardware you can manufacture at scale, what your models can actually do, and a big market need. In the United States, anything in the physical economy has to be transported on a pallet or in a cardboard box. So there’s an enormous amount of physical human handwork involved in transporting pallets and cases by hand.

Cassie is the result of that co-design. We manufacture it at very large scale and deploy it to a wide diversity of businesses across manufacturing and logistics. Our second robot is called Sunny. Sunny is an industrial semi-humanoid. It’s as if you were looking at a Unitree and said, “I need this robot to run for 3 to 10 years in a factory at human rates and reliability.” That leads to a slightly different set of design decisions.

Sunny is wheeled, not legged, because most indoor industrial settings have polished ground floors. It has a much higher payload capacity, can move much faster, and meets industrial collaborative safety standards that enable it to operate around humans. We have 100 Sunnys operating in Boston in our data factory, collecting data for internal use and advancing our research agenda. We also now have a growing fleet of Sunnys out in the real world doing more dexterous bimanual manipulation.

How broad do you see this product line of different robot form factors? Can you capture a big chunk of the useful market with 3 of them, or do you need 30?

It depends how you define the useful market. Where do you want to go? If you want to live inside manufacturing and logistics, Sunny and Cassie today can probably do 80% to 90% of the work. They’re designed for those cuts.

Maybe just describe what that work is, because I’m not sure it’s properly understood.

Most people have a mental picture of what the industrial base or manufacturing and logistics look like in the United States, and it comes from the TV show How It’s Made. I don’t know if either of you has seen it.

Yes, of course. Of course, yeah.

You see these Rube Goldbergian machines—

Yeah.

—with levers, and you don’t see any humans in those pictures, right? Somebody built a machine, and that is the factory.

Yeah.

That does exist. If it exists, they probably filmed it and put it on How It’s Made.

Right.

Because that’s the top 0.01% of American manufacturing and logistics. What’s below that? You walk into these million-square-foot facilities and see a bunch of people standing around conveyor belts or folding tables, doing things by hand. It’s kind of shocking for anybody who doesn’t come from this industry and community.

You look at a robot like Cassie, and there’s a huge amount of backbreaking work that’s really hard to hire for. There’s a 3-month turnover, and the average age of the worker doing that work is 55 and about to retire. We could use a Cassie to do that backbreaking work.

Similarly, Sunny can do basically anything you ever bought that came in a box. A human had to put that in a box, which is a crazy revelation for most people. Anything that requires any degree of assembly—even iPhones, where you’d assume a robot is putting the packaging together—is done by a person, even at the highest level of scale.

Then to your question: you have the single arm with the sucker—

Yeah.

—and the two-armed humanoid.

Yeah. What can’t they do?

Yeah, what can’t they do, and where do you go from there?

They can’t go outside, and outside is really interesting.

Is that just a locomotion issue?

It’s not locomotion; it’s the hardware. These robots have wheels and a little bit of traction. They have articulating hind legs, but they can’t drive over gravel. There are huge opportunities for robots in construction and for robots reclaiming materials from landfills. We need robots with legs for that. I think that’ll eventually be part of our pantheon of embodiments.

Similarly, you probably don’t want a Cassie or Sunny in your home just now. It’s a little too industrial of a device.

Yeah. They look scary.

Yeah.

Let’s talk about the customer adoption side. How has that been? Not strictly from an “I don’t have people, so I’ll take whatever I can get” perspective, but once they have a robot in their assembly line or wherever, how does adoption go in terms of their mindset?

2. Winning The First Robot Customers

It’s an interesting challenge because almost everybody we’re selling a robot to has never purchased one before. That comes out of these stats: if the United States is deploying only 38,000 robots a year, and 9 out of 10 U.S. factories and warehouses don’t have any robots, then almost all of the market is greenfield.

These people understand what the old guard of automation looks like, and they assume a robot doesn’t work for them. The way we break through that isn’t by talking about AI, models, Tutor, or anything like that. We call them up and say, “Have you ever considered a robot for $16 an hour?” They haven’t. That’s a new proposition for them. You can book a meeting, explain what we’re doing, and get people interested and involved.

By creating a business model with true incentive alignment between us as the robot vendor and the customer, we’re able to distribute robots at velocities many orders of magnitude faster than the traditional industrial automation industry. The average time from the first call to signing a robot contract is 40 to 60 days. From the signed contract to robots actually arriving at the customer’s site and doing work, it’s about 40 days. We stand up robots in production within a day, often within the first couple of hours that a robot is on-site.

We’re able to move very, very, very quickly because the robot is a generally intelligent, generally capable machine that we’re not customizing. We’re just shipping it directly from the factory to the customer. And we’re not putting any economic risk on the customer, so we’re able to get them on board.

And once they’re deployed, how do people’s attitudes about labor versus machines change?

The operations community is really fantastic because they see something work, and it’s like, “It works.” That’s it. They can see that it does the job. They know all the places where this job needs to be done, so let’s put it in all those places. We’ve been able to scale up very quickly, both with small businesses and with some of the largest companies in the supply chain, going from one robot at one site to a dozen-plus robots across many sites.

It is interesting, though, because with a lot of Gen AI, there’s a social element: people get very attached to their models, and it’s hard to swap models.

Yeah.

Listen, you probably have more experience deploying robots than most people alive at this point. Are you seeing the same thing? Are they viewed as equipment, or are they starting to be viewed as beings and coworkers?

I saw a very funny tweet: “A robot is a robot unless it works, in which case it’s a machine.” I’ve got to figure out the attribution.

But is it real?

It’s pretty real.

Is that what it is? They just view it as a piece of equipment once it works?

Certainly at the factory manager level, yes.

Oh, okay.

Interestingly, in robotics, your buyer is not your user. Your buyer is the COO or the general manager, and the user is a factory worker—

Yeah.

—who’s working with the robot all day long. There, you do need to cultivate the relationship between those workers and the robots because we don’t fully automate 100% of a factory. We automate some jobs in a factory or warehouse.

Do you have a philosophy on how you approach this? Do you try—

Oh, yeah.

—to make them look friendly? Do you—

Yeah.

—make them look more like coworkers or less like coworkers? Do you make them talk? How do you view this philosophically?

Very luckily, there’s a whole school of academic research called HRI, or Human-Robot Interaction.

Right. Had no idea.

There you go.

A lot of people have spent a lot of time thinking about this, including a bunch of people from my academic background.

Okay.

There are a whole bunch of tricks. One is that every robot has a name, which lets us build a relationship with those robots.

Who names it?

Mostly our manufacturing technicians.

So you name it, not the customer? It’s not like with a dog, where when you get a new dog, you can rename it.

When a customer expands with Tutor and places a repeat order, they’re allowed to name the robots.

Oh, okay.

Oh, okay.

Because you’re not supposed to rename a boat—

Yeah, yeah, yeah.

—because it brings bad luck, so you have to keep the boat name. But you can rename a dog, so what—

The new robots get to be named.

Got it.

That solves it, and we get customers who get very creative with this.

They have a higher sense of ownership now that they’ve named it.

Exactly. We get really strong signals. We’ve had people name the robots after their kids.

Whoa.

Whoa.

People name the robots after their favorite football player or their favorite pop star. People get very attached because the robot ultimately helps. It makes their jobs easier. If you’re a line worker in the facility, you’re not doing as much backbreaking work. You’re able to move into a higher position, and none of these places can hire. If you’re a factory manager, your job is a thousand times easier.

What naming scheme do you use when you release them?

We started off naming them after hurricanes, and then—

So that’s not—

Yeah.

You didn’t read the HRI book, clearly.

Well, you’re an academic.

The customer doesn’t know that Britney is N mod 26 or whatever equals 2. We actually stopped because we went through the hurricane names—we went through the alphabet too many times to keep track of it.

Okay. So what are you using now?

I think the technicians just pick whatever brings them joy. We have the wonderful privilege of employing a big staff of people in the United States who are helping to build these robots.

Amazing.

And they’re also naming them.

Awesome.

The other thing that surprised me when we first did the work before investing was that most of the time, when we talk to robotics companies, they’re going after these humongous companies—the big Fortune 50-type companies that seem to have the budget for lots of experiments.

Yeah.

Whereas I think a significant majority of your customers are these local businesses.

Yeah.

Is there a message in that? Was it a conscious strategy?

It’s a lot of things. One is that’s most of the market. The vast majority of U.S. businesses in manufacturing and logistics are small or mid-market businesses. If you want to serve the market, that’s what it takes.

I think we also started this company at a time when we were seeing all these very academic, supply-chain-oriented robotics startups get started and then not really succeed. We had the opportunity to ask, “Okay, what went wrong here?” One of the big things robotics companies get wrong is that they’re not really product companies. They say, “Okay, some really big company has a very specific problem. They’re willing to write us a shiny piece of paper that says they’re going to expand with it. We’re going to spend 2 years building a very custom system to solve their problem. We’ll deploy it.”

Chances are it won’t work because, as anybody who’s built products knows, you can’t just one-and-done a product. Doing product requires iteration. And even if it does work, who’s to say that customer still has that problem? The company is done. I saw this happen five to 10 times in a row, to the point where I thought, “Oh my God, if this happened to us, that would just be embarrassing.”

Working with a diverse range of customers is a beautiful forcing function for building the right kind of product, which to us is a general robot. We don’t want to build a Rube Goldberg machine that works really well for big industrial customer A or B. We want to build a platform with generally capable robot bodies running generally capable robot models that can immediately get to work. If you can sign a contract in 2 weeks with a mid-market customer and have a robot running immediately after that, you know you’re building a totally different kind of product and business.

How do you think about what the industry looks like in the limit, assuming you continue to be successful? Is it just that this work gets cheaper faster, or does quality somehow go up? Can they do more things? What’s the economic convergence or the productivity convergence?

I think everything gets better. There’s a whole set of goods that, if they’re cheaper, we’ll buy more of them. If you believe those goods are valuable, that’s good for everybody.

There are also a bunch of discontinuities in the world. I was talking with somebody the other day who worked on the line in a housing factory in, I think, Wisconsin. They were describing how the houses they’re building are about 5% more expensive to build in a factory than through traditional on-site construction.

But if these houses were even just 1% cheaper than the equivalent-quality house manufactured or constructed on-site in the traditional way, all the demand from developers would immediately go to that other way of building these homes in a factory. The same thing could be said about reclaiming materials in landfills and a whole bunch of other markets. So I think our hope is that if Tutor is really successful, we can provide this robot substrate, this labor force that other businesses can use to transform the physical economy and make the world ultimately a cheaper and better place to live in.

Will you ever go in the home?

Yeah, we’re not opinionated. We’ll go anywhere.

Yep. But I would imagine that would be a much tougher technical problem, just from—

Yeah.

—the complexity of the task and safety—

And the economics of it, right? What is the economic value a Tutor robot delivers in a factory, and how many consumers are willing to pay for that in their home? It doesn’t quite work out. But again, that comes back to our overall philosophy: take trade-offs across as many layers of the stack as possible to get a foothold, start building the platform, and as you build the platform, you have to take fewer and fewer trade-offs over time.

How does product innovation work in robotics? Most companies are just on V1, whereas you’ve got 2 versions of the robot and continue to iterate really fast. What’s different? How is it different from digital product innovation or other categories?

3. Making Robotics A Product

We’re trying to make it as similar as possible. We’re manufacturing our own robots. This gives us the ability to own that manufacturing process and figure out how to push changes to it every 2 weeks. We’re shipping code to robots either every day or in major releases every 2 weeks, and we’ve built a huge amount of infrastructure to support that level of iteration.

We’ve built, I think, a really unique shape of product function, where our product function in the organization sits right next to our customer success and implementation functions—

Mm-hmm.

—and they operate as one big team. So they’re as close to the customer as humanly possible, which I think has been really successful for us. We also have a research team that’s able to be both deeply embedded in the customer problem space—we’re able to pull all sorts of levers to get specific types of data from customers—and take these moonshots. If we have a breakthrough on our modeling side that lets robots do a whole new shape of thing, we can co-design a product around that in a way that I think most companies couldn’t, and get it out into the field much quicker.

If I’m listening to this and I have a warehouse—

Yes.

—and I want to get a robot, do I just go to the website and put in my email? How does this work?

That’s it. You can go to our website and put in your email. We actually have cool pricing calculators where you can put in some operational data about how your warehouse or factory runs today.

Okay.

And you can get an estimate: with an initial deployment of Tutor robots, we could be saving $100,000 a year, $1 million a year, or $100 million a year. And that’s not, “Okay, this is a roadmap over 5 years of custom engineering.” It’s like, no, you could save $1 million a year 2 months from today, without putting CapEx down.

Have you found that there is demand? It’s just such a sci-fi thing. Is there demand out there?

There is so much demand, and it’s not even demand for robots, because people don’t know that this shape of robot exists, or that it works, or that it can do this job. Ultimately, we have this framework: the value of a robot is how much work a robot can do divided by its cost. That’s it.

Yeah.

Right? Everything else is commentary.

Sure.

And so if we’re able to compete in these highly liquid labor markets that exist across the American physical economy, and we can offer something that delivers greater equivalent capability at a lower cost, the market for that is the whole market, right? It’s all physical work done.

And just one quick wonky question on this: when you say cost, how much do you mean amortizing the upfront cost versus the ongoing cost to operate it?

It’s both, which is very interesting and different, right?

Yeah.

Traditionally, we’ve thought about—

Yeah.

—robotics as ROI, right?

Yeah.

Big CapEx expense and then 10 years to make it back. Now, more and more of the cost is going into ongoing inference and teleoperation.

But does that also include maintenance—

Yes.

—and break-fix and all of that stuff?

It has to.

And what is the dominant cost? Are they roughly equal?

We drive down all of them over time.

No, I’m just saying from your standpoint—

Yeah.

—when you pencil out your COGS, how much of it is the upfront cost versus the ongoing cost?

It is more ongoing at this point, and that’s surprising to people.

Interesting.

Oh, wow.

Yeah.

Yeah.

Not surprising to me. You’re really competing with something that you feed rice to, and it’s very, very efficient, right? The human body.

So when you get customers, is it because Black Friday or something is coming up and they need a bunch of units done in a really short time, or is this actually now part of their—

Yeah.

—labor planning cycle?

No, it’s strategic. These businesses are faced with a really hard choice: they know the age of their workforce, they know that demand is ramping up due to all these secular trends, and the cost basis of fulfilling that work is going up. They’re asking, “How does our business pencil out?” These aren’t software businesses with huge gross margins that you can play with and compress a little bit, and it’s fine. It’s like, no, the business either works or it doesn’t.

I think our customers are ultimately looking for a long-term partner that can do the robots, right? If one of our customers is manufacturing deodorant, they’re doing the deodorant, right? They’re going to be operational experts in deodorant, or operational experts in crayons, or whatever the thing they manufacture is. They need a partner who’s the expert on robots.

I have a question on—

Great.

—the nature of factories. You can ship a robot to anybody anytime—not today, because of your shortages, but at some point in the future, right? And if your robots become more and more general-purpose over time, what does it mean for where factories are located and how people organize around these factories and these one-factory towns, whatever the expression is?

Yeah.

All of that fundamentally changes now, right?

Yeah. I think a lot changes. There’s a lot of geopolitics in this. I think one big thing is this is going to be a great equalizer between the United States and—

Yep.

—or excuse me, other countries like China that invested in the previous regime of industrial automation. In some ways, that made them very successful, and in other ways, in hindsight, that might end up being an enormous mistake. In the United States, we’re doing this greenfield build of automating our industrial capacity using generally capable robots. In China, they’re going to have to tear down. We’re not going to have to do that in the United States in the same way.

I think another thing that’s going to shift, which is really interesting, is we’re going to have way more factories, in the same way that we have way more businesses because the cloud and all of the software tooling let people build new businesses. Anybody will be able to start a business in the physical world. Anybody will be able to scale a business in the physical world if we have this infinite, flexible labor substrate.

You brought up China. We actually published a report on the supply chain in robotics, and it seemed that China was so central to it.

Yeah.

So can you talk a little bit about your dependency on it, how that evolves over time, and how much risk there is in China to the supply chain?

Ultimately, no matter what anybody says, today in 2026, we have a highly interconnected, interdependent global supply chain, and this is one of many examples of that.

Yeah.

I think both U.S. companies and Chinese companies are thinking very hard about what the long-term strategy of this looks like. It’s Chinese companies investing in the United States, it’s U.S. companies investing in China, and ultimately everybody trying to build their industrial capacity.

Yeah. I’ve noticed that you’ve verticalized more, as far as I can tell. Is that right?

Yes.

Like, just taking more in-house. And do you plan to continue doing so?

Yeah. Well, there’s this tough thing in the supply chain, especially because robotics is such a cottage industry. You might start in a world where subcomponents are coming from OEMs, and then you grow your volume to the point where they’re not OEMs anymore. They’re glorified contract manufacturers, right, once you’re doing a material percentage of their volume. At that point, the relationship and the dynamics change. It’s like, is that the right command for that thing? Do I want to be using a command, or do I want to be manufacturing it in-house?

Yeah. I mean, Tutor must be one of the largest deployers of robots in the United States. As a result, you must be one of the largest customers for some of these OEMs. I would imagine that over time, you just continue to drive these components yourself.

I think that’s right.

Does that mean taking it in-house or just telling them what to do? Are they still suppliers?

Some combination, I guess.

Is it?

Some combination. I think ultimately, for Tutor to be successful, we’ll need a mature robotic supply chain in the United States.

Yeah.

We’re in this position where, because we’re so much of the demand for that supply chain, I think it’s both our obligation and our opportunity to invest very, very, very heavily in building this capacity in the United States. That’s why we build robots in Watertown, Massachusetts. It’s why we expect to continue to verticalize and build more and more of our robot componentry in the United States.

Again, going back to when we first started talking, I remember it was probably the second meeting when we actually started talking about the research, right? Most of the meeting was about the business because we had such a great business. But that’s a big strength of yours, right? Talk about how you approach it. There are all sorts of theories about which robotic learning models will win, and so on and so forth. How are you approaching the learning bit—

Yeah.

—in research?

I think we’re thinking about how this played out in LLMs and what’s similar and what’s different. One thing that’s clearly been very, very important in LLMs is that there are lots of heterogeneous sources of data, but ultimately there’s a top of the data pyramid, which is real-world reinforcement learning traces of doing the thing you want to be able to do.

We’ve built our research program around how we get the most out of robot experience in the real world. We built this 100-robot fleet called Data Factory One—

Yeah.

—in Boston. This is, to our knowledge, the largest semi-humanoid data collection farm in the United States, and we’re able to use it as a research test bed for how to do real-world reinforcement learning at scale with robots. Reinforcement learning is a family of techniques that doesn’t really work at small batch sizes. If you have a very large fleet out in the real world, you can start to play with these methods, which I think people haven’t been able to do either in academia or in industry.

So I presume building a team for this is quite different than a lab. Maybe talk about the construction of the team, and whether there’s anybody you’re looking to hire.

We’re looking to hire everybody.

Everybody.

We’re looking to hire researchers and engineers across every single layer of the stack.

Wow.

We’re also looking to hire technicians to build robots.

By the way, when you say every layer of the stack, you mean software—

Hardware—

—hardware—

Materials—

—electronics—

—electronics.

Materials, everything.

There’s nowhere that you can abdicate responsibility, right?

It’s robotics.

You’re going to engage with it whether you want to or not. The question is whether you understand it. We’re building that team, and I think that’s a really unique shape of team, not only because it’s so interdisciplinary and collaborative across all those layers of the technical stack—

Yeah.

—but because I think it’s one of the only places in the universe where you can be at a whiteboard talking about model architecture in the morning, and in the afternoon, looking at a factory line, seeing how a robot is doing, and talking.

Is this primarily in Massachusetts?

We have a team across the United States.

Okay.

We have folks on go-to-market and technicians who are engaging with our customers. Our customers are everywhere across the United States.

That’s great. All right. Awesome, dude.

Thank you.