The Robots Already Working in American Factories | Tutor Intelligence

1. Key Themes

Robotics Is the Ultimate Systems Problem, and Every Layer You Skip Costs You 90% of Your Tradeoff Space

Josh Gruenstein's core thesis is that robotics cannot be won at a single layer (model, hardware, or customer). The companies that fail are those that refuse to touch parts of the stack. Raghu Raghuram frames this as robotics being "the ultimate systems company," and Josh goes further: robotics isn't just model/hardware/software, it includes supply chain, customers, product, and business model.

Josh explains: "If you view robotics in the same way that I do, that it's this, like, ultimate systems problem, the only way you can make forward progress is by figuring out all of the clever tradeoffs across as many layers of the stack as you can." [00:05:48] He adds: "If you throw out one of the layers of the stack and you're like, 'Well, we're not going to touch that,' you've eliminated 90% of your trade-off space to make progress." [00:06:14] Martin Casado's observation that "most robot companies start horizontal, and they all go vertical" [00:07:09] is validated by Tutor's trajectory, from buying components to building robots in Watertown, Massachusetts.

Deploy First, Let the Data Be an "Emission Byproduct" of Economic Work

Tutor's central strategic insight is to get data not by paying for it but as exhaust from robots doing paid work. This guarantees the data is economically valuable, which is something synthetic or staged data collection cannot promise. The name "Tutor Intelligence" refers to human teleoperators ("tutors") who teach robots new tasks while they work.

Josh describes the model: "What if we could build this network of robots out in the real world, we could connect them with human teleoperators, which we call tutors, that teach the robots how to do new tasks, and in the same way the internet kind of, by emission, created this data set to train these frontier models, we can do the same with robots." [00:10:07] On the economic alignment: "If you have robots in real customer sites that customers are paying for for those robots to do that task, you know for sure that that data is economically valuable." [00:14:12] And contrasting with alternatives: "If you're like, 'Okay, we're just going to strap cameras on everybody's heads and send them throughout Indonesia to collect data doing XYZ,' you have no guarantee." [00:13:54]

The Market Is 90% Greenfield, and the Sales Pitch Is Labor Cost, Not AI

Almost no US factory or warehouse has ever bought a robot, so Tutor is not displacing incumbent automation. It is selling to a market with a mental model that "robots don't work for me." The unlock is selling a price per hour rather than technology.

Josh: "9 out of 10 US factories and warehouses don't have any robots, then almost all of the market is greenfield. And these people, you know, understand what this old guard of automation looks like, and they assume a robot doesn't work for them." [00:23:51] The pitch: "We cold call them up, and we're saying, 'Hey, have you ever considered a robot for 16 bucks an hour?' And they haven't." [00:24:12] The resulting velocity: "We have an average time from that first call to a robot signing contract of 40 to 60 days. We have an average time from that signed contract to robots actually in the customer site of about 40 days doing work." [00:24:32]

Robots-as-a-Service Economics: OpEx-Dominated, Not CapEx-Dominated

A counterintuitive finding is that Tutor's cost of serving is now mostly ongoing (inference, teleoperation, maintenance) rather than upfront hardware. This reframes robotics from a "big CapEx, 10-year ROI" business into something resembling a cloud services business with a utilization profile.

Josh: "Traditionally, we've thought about robotics as ROI, right? Big CapEx expense and then 10 years to make it back. And now it's actually like, no, more and more of the cost is going into ongoing inference, teleoperation." [00:36:38] Asked about COGS: "Yeah, it is more ongoing at this point, and that's surprising for people." [00:37:11] He applies cloud-style thinking to humans-in-the-loop: "Think about what is the cost to serve models with an H200... you can think very similarly about human intelligence." [00:11:57] The value framework is simple: "The value of a robot is you take how much work a robot can do and you divide it by its cost. And that's it. Everything else is is commentary." [00:35:52]

Frontier Labs' Spatial Reasoning Investments Are an Accelerant for Robotics

Rather than viewing frontier models as a threat, Tutor treats them as a free input. Labs are post-training on 3D spatial data (CAD, 3D modeling) because it's monetizable, and that happens to be exactly the capability robotics lacks.

Josh: "The labs are now post-training their models on 3D spatial data like CAD and, like, 3D modeling, which by the way they're doing because there's this direct economic incentive, because there's a real market for it. So, they can immediately monetize it. And it turns out that's the exact spatial reasoning that our models need. So, I think we've been catapulted many years ahead in the future." [00:15:49] Raghu notes the systems framing means the effect is "acceleratory as opposed to slowing you down." [00:16:24] Josh's durable moat is the flywheel: "The thing that has kind of the long-standing generational value in a business is the flywheel and loops themselves." [00:16:46]

Product Velocity in Hardware: Software-Style Iteration Cadence

Tutor attempts to mimic SaaS release cadence in physical products by owning manufacturing and co-locating product with customer success.

Josh: "We are shipping code to robots either every day or major releases every two weeks and have built a huge amount of infrastructure to support that level of iteration." [00:33:59] On manufacturing: "We're manufacturing our own robots. This gives us the ability to own that manufacturing process and figure out how do we push changes to the manufacturing process every two weeks." [00:33:59] Raghu's observation: "Most companies are just on V1, whereas you've got two versions of the robot, and you continue to iterate really fast." [00:00:45]

Verticalization Driven by Supply Chain Scale and the "OEM Becomes a Glorified Contract Manufacturer" Dynamic

As Tutor scales, it becomes the dominant demand source for a supply chain that barely exists in the US. The relationship with component suppliers changes, forcing in-house build or deeper control.

Josh: "You might start in a world where subcomponents are coming from OEMs, and then you grow your volume to the point of they're not OEMs anymore, they're glorified contract manufacturers, right, once you're doing a material percentage of their volume." [00:41:01] And the strategic imperative: "For Tutor to be successful, we will need a mature robotics supply chain in the United States... because we are 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 to building this capacity in the United States." [00:41:49]

Human-Robot Interaction as a Go-to-Market Lever

The buyer (COO/GM) and the user (line worker) are different people in robotics, so adoption requires cultivating worker affinity. Tutor uses naming and ownership to build it.

Josh: "In robotics, your buyer is not your user, right? Your buyer is like the COO or the general manager, and the user is like a factory worker, right, who's working with the robot all day long. And there you actually do need to cultivate the relationship between those workers and the robots." [00:26:42] The tactic: "When a customer expands with Tutor, when they do a repeat order, then they're allowed to name the robots." [00:27:48] and "People name the robots after their kids... their favorite football player, their favorite pop star." [00:28:06]

Robots as a "Labor Substrate" That Multiplies the Number of Physical Businesses

Josh's long-term vision is the robot as cloud-equivalent for atoms: a flexible labor layer that lets anyone start or scale a physical-world business, producing more factories and geopolitical rebalancing.

Josh: "In the same way that we have way more businesses because kind of 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 kind of have this infinite, flexible labor substrate." [00:39:35] And: "Our hope is like if Tutor is really successful, we can provide this robot substrate, this labor force that other businesses can use to transform the physical economy." [00:32:42]

2. Contrarian Perspectives

The US May Benefit From Having Under-Automated Factories

The conventional view is that China's head start in industrial automation is a durable advantage. Josh argues the opposite: the old generation of automation is a liability once general-purpose robots arrive, because it must be torn down.

Josh: "This is going to be a great equalizer between the United States and other countries like China invested in the previous regime of industrial automation. And in some ways, that made them very successful, and in other ways, I think, in hindsight, that might end up being an enormous mistake because, 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." [00:39:05] The supporting statistic: the US deployed just "38,000 industrial robot workcells" in 2025. [00:17:27]

Winning Robotics Startups Should Sell to Mom-and-Pops, Not Fortune 50s

Most robotics companies chase giant enterprises with big budgets for experiments. Josh argues this is precisely the trap: it produces custom, one-off systems and existential customer concentration.

Josh describes the failure mode: "They're like, '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 two years building a very custom system to solve their problem. We'll deploy it. Chances are it won't work... And then even if it does work, who's to say that customer still has that problem?' And so boom, the company's done." [00:30:16] He saw this "5 to 10 times in a row." [00:30:51] A diverse customer base is "a beautiful forcing function for building the right shape of product, which, to us, is a general robot." [00:30:58]

Don't Lead With AI: The Best Robot Is the One Customers Don't Notice

Counter to the industry's obsession with showy demos, Josh argues that the signal of a good robot is that customers talk about shampoo cost, not the robot. Raghu noted pre-investment diligence calls revealed customers neither knew nor cared when a human tutor versus the model was operating.

Josh: "You want the feedback to be like, 'My shampoo that I'm manufacturing at volume is now 5 cents cheaper than it was.' If the feedback is like, 'The robot is pretty,' it's like, 'Ah, it's probably a really bad robot.'" [00:11:37] And on sales: "The way we break through that actually isn't by talking about AI or models or tutors or anything like that." [00:24:12]

Data Strategy Should Follow Economic Incentive, Not Academic Purity

In a field enamored with world models, human video, and simulation, Josh insists real-world RL traces from paid work sit at "the top of the data period," with other sources as supplements.

Josh: "There are lots of heterogeneous sources of data, but ultimately, there's a top of the data period, which is real-world reinforcement learning traces doing the thing you want to be able to do." [00:42:43] And: "We're not purists. Again, we're willing to take any trade off, right? We love using simulation to kind of widen these data distributions." [00:14:47]

Discontinuities, Not Gradual Savings, Are the Real Prize

Josh argues that small cost reductions can trigger step-change demand shifts in industries, citing prefab housing.

Josh: "The houses that they're building are about 5% more expensive to build them in a factory than to do traditional construction on-site. But if these houses were even just 1% cheaper than kind of the equivalent quality of house manufactured or constructed on site in the traditional way, all of the demand from developers would immediately go to that other shape of building these homes in a factory." [00:32:07]

3. Companies Identified

Tutor Intelligence. Boston-area robotics company (~160 people) building general-purpose robots, training frontier robot models, and manufacturing robots in Watertown, MA; deployed across small businesses to Fortune 50 customers. Raghu and Martin discuss it as one of the few robotics companies actually deployed at scale. Josh: "Tutor robots have helped manufacture and ship over a quarter of a billion products to American consumers." [00:00:24] Martin: "You're one of the few companies that's actually deployed robotics at scale." [00:02:38]

Unitree. Chinese humanoid and quadruped robot manufacturer. Referenced as the reference point for off-the-shelf humanoids. Josh notes Tutor's second robot is designed as if "you were looking at a Unitree, and you were like, 'I need this robot to run for 3 to 10 years in a factory at human rates and reliability.'" [00:19:57] He also notes Unitrees were "probably zero" among 2025 US industrial deployments. [00:17:36]

FANUC. Traditional industrial robot manufacturer representing the "old guard" of automation. Josh: "The robots that FANUC and ABB are building, they're mostly not building them in the United States... they're building these objects that are designed for systems integrators that are designed for fast repetitive motions. They are not cut from the set of tradeoffs that you would want to co-design with frontier models." [00:17:38]

ABB. Industrial robotics incumbent, mentioned alongside FANUC as building for systems integrators rather than for co-design with AI models. [00:17:36]

a16z. Investor in Tutor Intelligence; Raghu and Martin referenced diligence work, including calls with customers and a published report on robotics supply chains. Martin: "We actually published a report on the supply chain in robotics, and it seemed to be that China was so central to it." [00:40:02]

Anthropic's Claude. Used by Josh to estimate Tutor's impact. Josh: "Claude told me last night that it estimates Tutor robots have helped manufacture and ship over a quarter of a billion products." [00:02:43]

NVIDIA (H200). Mentioned as the analogy for cost-to-serve and utilization economics of tutors versus models. Josh: "What is the cost to serve models with an H200, right? And you can map that out, and what's the duty cycle and utilization of that H200?" [00:11:57]

GPT models / frontier labs. Referenced as holding general world knowledge that robotics couldn't previously express; labs are now post-training on 3D data. Josh: "Everybody knew that all of this general knowledge about the world was baked into these models, but they couldn't really express it." [00:15:24]

How It's Made (TV show). Used as the mental model of top-tier automated factories; Josh argues those represent "the top 0.01% of American manufacturing and logistics." [00:21:37]

Astra (as transcribed). Raghu references "models like Astra" having impact on simulation; the term as transcribed is unclear and is noted only as a reference to emerging spatial/world models. [00:15:17]

4. People Identified

Josh Gruenstein. Co-founder and CEO of Tutor Intelligence; MIT robotics background, full-stack roboticist turned operator. He is the primary subject. Martin: "Whenever I talk to you, I feel like I'm talking to an operations guy that talks about unit economics and operations and supply chain." [00:05:25] Raghu adds "And even a go-to-market guy." [00:05:35] Josh on his own strength: "My strength as a roboticist was that it was very full-stack. And I was willing to think about the manufacturing, the hardware. I was willing to think about FPGAs and ASICs." [00:05:48]

Martin Casado. a16z general partner and co-host; probes the horizontal-vs-vertical question and the economics of tutors. Not discussed for excellence but notable for the sharp framing: "Most robot companies start horizontal, and they all go vertical." [00:07:09]

Raghu Raghuram. a16z partner and co-host; describes the diligence that led to the investment. Raghu: "When we talked to your customers before we invested, most of them didn't seem to know or didn't even seem to care that there was a tutor or when the tutor was operating and when your learning model was operating." [00:11:15]

Tutor's manufacturing technicians. Unnamed Watertown technicians who build and name the robots. Josh: "We have a really wonderful and privileged to employ a big staff of people in the United States who are helping to build these robots, and they are also naming them." [00:29:06]

Unnamed Wisconsin housing factory worker. Source of the prefab housing cost insight. Josh: "I was talking with somebody the other day who worked on the line in a housing factory in I think Wisconsin." [00:32:07]

Unnamed tweet author. Source of the line "A robot is a robot unless it works, in which case it's a machine." Josh: "I got to figure out the attribution." [00:26:36]

5. Operating Insights

Make the Customer's Buyer Frame a Price per Hour, Not a Technology Purchase

Tutor reframes the robot as a labor substitute priced against an hourly wage, which collapses sales cycles because the buyer already has a benchmark. Josh: "We cold call them up, and we're saying, 'Hey, have you ever considered a robot for 16 bucks an hour?'" [00:24:12] Combined with no upfront CapEx: "We're not putting any sort of economic risk on the customer. So, we're able to get them on board." [00:25:12] The pricing calculator extends this: "You could save that a million dollars a year 2 months from today without putting CapEx down." [00:35:25]

Embed Product Next to Customer Success and Implementation as One Team

Rather than the typical separation, Tutor collapses product, customer success, and implementation to reduce the loop between customer pain and product change. Josh: "Our product function in our organization actually sits right next to our customer success and implementation functions, and operate actually as one big team, so they're as close to the customer as humanly possible." [00:34:19]

Research Team Embedded in the Customer Problem Space, With Room for Moonshots

Tutor's research org is both customer-driven and free to co-design products around breakthroughs. Josh: "We're able to pull all sorts of levers to get specific types of data from customers, but also able to take these moonshots. And, you know, if we have a breakthrough on our modeling side that lets robots do a whole new shape of thing, we can kind of co-design a product around that." [00:34:29]

Build a Dedicated Data Factory to Make RL Work at Batch Sizes Academia Can't Reach

Real-world RL needs scale. Tutor built a 100-robot fleet purely for research. Josh: "Reinforcement learning is a technique family where it doesn't really work at small batch sizes. And so 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 play with either in academia or industry." [00:43:11]

Use Ownership Mechanics to Create Adoption Among End Users

Tying the naming privilege to repeat orders turns a cosmetic feature into an expansion incentive and a retention signal. Josh: "When a customer expands with Tutor, when they do a repeat order, then they're allowed to name the robots." [00:27:48] Raghu: "So they have a higher sense of ownership now that they've named it." [00:28:04]

6. Overlooked Insights

Operational Time-to-Production Is Measured in Hours, Which Redefines Robotics as a Product Category

Dropped in the middle of the customer adoption discussion was a line that is easy to skim past but is a radical break from industrial automation norms (which measure deployment in months or years): "We stand up robots in production within a day, often within the first couple of hours that a robot is on site." [00:24:32] This is only possible because "the robot is this generally intelligent, generally capable machine that we're not customizing, we're just, you know, shipping directly from the factory to the customer." [00:25:12] Combined with a roughly 40-day contract-to-deployment window, this implies Tutor has effectively productized deployment, which is the true moat against systems-integrator-based competitors, whose business models depend on customization hours.

The Labor Market Dynamics Make This a Demographic Trade, Not Just a Cost Trade

Josh mentions in passing that the work Tutor's robots do has "a three-month turnover, where your average age of the worker doing that work is 55 years old and about to retire" [00:22:02], and later that "these businesses... 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." [00:37:41] And critically: "None of these places can hire." [00:28:21] This reframes the addressable market: it is not only substitution of existing labor at a lower price, but fulfillment of labor that does not exist. That implies demand is less price-elastic than typical automation and that the real competitor is unfilled capacity, not the incumbent worker, which also helps explain why operations leaders adopt so quickly once they see the robot work.