In this episode of 'The Next Next,' host Jason Jacobs interviews Jon Miller Schwartz, co-founder and CEO of Ultra. Ultra is a company focused on developing practical, general-purpose robots designed to adapt to various tasks and environments in American warehouses. The discussion covers Ultra's mission to revolutionize industrial automation, the significance of AI advancements in making adaptable robots feasible, and Jon's background and entrepreneurial journey. They delve into the company's strategic focus on the e-commerce logistics sector, the operational challenges they face, and the importance of data in refining robotic capabilities. The podcast also touches upon the broader implications of these technological advancements for labor and society. The conversation reflects on current and future goals while underscoring the collaborative efforts within the industry to drive progress.
Revolutionizing Warehouse Automation with Jon Miller Schwartz of Ultra
In this episode of The Next Next, host Jason Jacobs welcomes Jon Miller Schwartz, co-founder and CEO of Ultra. Ultra is pioneering the development and deployment of general-purpose robots aimed at transforming American warehouses. Jon discusses Ultra's mission, the importance of adaptable robots versus traditional static systems, and the role of AI in making these advancements possible. He outlines Ultra's unique technology, the challenges they've faced in the robotics industry, and their strategic focus on the e-commerce logistics space. The conversation also delves into the benefits of these robots for customers, the current state and future of robotics, and the importance of data in improving robotic capabilities. Jon shares insights about Ultra's rapid growth, their operational strategies, and the immense potential for robotics in various industries.
00:00 Introduction to The Next Next
02:01 Meet Jon Miller Schwartz: Ultra's Mission and Background
02:43 The Evolution of Robotics and AI
05:24 Jon's Journey: From 3D Printing to Robotics
07:30 Challenges and Innovations in Robotics
14:51 Focusing on Logistics: Ultra's Strategic Direction
20:07 Building and Deploying Robots: Ultra's Progress
22:34 Targeting the Logistics Industry: Ultra's Customers
26:22 Augmentation Strategy and Industry Growth
26:50 Robots in Fulfillment and Labor Optimization
28:03 Robot Performance vs. Human Packer
29:29 Cost Efficiency and Utilization of Robots
30:16 Impact on Human Workers and Job Shifts
32:31 Challenges in Robotics and Future Prospects
34:59 Focus on E-commerce Fulfillment
36:24 Competition and Defensibility in Robotics
37:42 Data Collection and Model Training
44:13 Technological and Societal Readiness
48:55 Future Vision and Company Goals
50:50 Hiring and Collaboration Opportunities
51:49 Conclusion and Final Thoughts
Jason Jacobs: Today on the next. Next, our guest is Jon Miller Schwartz, co-founder and CEO of Ultra Ultra builds and deploys practical, general purpose robots for American warehouses. We cover a lot in this episode, including Ultra's mission to revolutionize industrial automation, specifically focusing on robots that can adapt to various tasks and environments without requiring extensive re-engineering.
He also discusses the background of the company, their operational model, the progress that's been happening generally in AI and robotics, as well as ultra's progress to date. We talk about their strategic focus on the e-commerce logistics space and why we talk about some of the current projects that they've been working on and how they are interacting with their customers and involving them in the building process.
And we also talk about the importance of data in [00:01:00] refining robotic capabilities and the collaborative efforts within the industry to accelerate advancements. It's a great one, especially for me, who didn't have a lot of background in robotics. Coming into the show and I'm excited for you to listen, but before we get started.
I'm Jason Jacobs, and this is The Next Next. It's not really a show, it's more of a learning journey to explore how founders can build ambitious companies while being present for family and not compromising flexibility and control, and also how emerging AI tools can assist with that. Each week we bring on guests who are at the tip of the spear on redefining how ambitious companies get built, and selfishly the goal is for this to help me better understand how to do that myself.
While bringing all of you along for the ride, not sure where this is gonna go, but it's gonna be fun.
[00:02:00] Okay. Jon Miller Schwartz, welcome to the show.
Jon Miller Schwartz: Thanks, Jason. Glad to be here.
Jason Jacobs: Well, glad to have you. It's funny we got introduced through Derek Haswell, who we both know, and the introduction was timely because every now and again in my updates, I talk about robotics as an area that I am looking to understand better. And here you are a founder that is, smack in the middle of what's happening in the trenches with robotics.
So I'm psyched to have you on and learn more about Ultra and about robotics more generally.
Jon Miller Schwartz: Yeah, absolutely. Excited to chat about it.
Jason Jacobs: To kick things off, maybe just talk a bit about Ultra and what it is.
Jon Miller Schwartz: Sure. At Ultra, we are building what I like to call practical, general purpose robots. So we're focused on building robots that solve one specific problem today fully, but over time, we can use to solve other tasks, other problems with the same exact [00:03:00] robot. This is fundamentally different from what robots have historically looked like.
When you walk into most factories or warehouses, you'll see kind of industrial automation and those types of systems are, usually very customized to the specific, warehouse or the task that it's doing. And so every time you wanna make a change or something changes in the environment or the the items that the robot's interacting with, you need to call 'em the engineers and they need to go and re-engineer the robot.
Either the hardware, the software, or both. But we're now living through this moment. Due to a lot of the progress in AI where we can actually build robots that can constantly adapt and be flexible to changes in environment or task or the objects that they're manipulating. And so it's super exciting to be working in robotics today because we can really see this opportunity to go and scale the number of robots that are out there in the world.
Jason Jacobs: What is it about AI that that makes that possible?
Jon Miller Schwartz: Yeah, so you know. All of this is really based on the same [00:04:00] underlying innovations that have enabled, chat, GPT and other LLMs. To get a little technical, basically the transformer neural net architecture, which was pioneered at Google and OpenAI then realized could be applied to language to create the LLM.
It basically takes in language, so text from the internet and breaks it down and from there is able to predict, the next part of a word, or, eventually if you play that out. Many words, many sentences, many paragraphs. And so that is, how an LLM works. And if you think about that, if you swap out the text for, let's say robotic data, so the positions of a robotic arm over time you can actually apply that same type of like neural net prediction to robotics.
And so instead of predicting language, we're actually predicting the position of a robot and what it should be as it, interacts with the world. And so the inputs, maybe rather than a prompt, are a camera feed or tactile sensors, or in some cases [00:05:00] a prompt, like a task that you want a robot to do.
And so this basically means that we can now have robots that are trained with data instead of needing to be explicitly programmed by, engineers who have robotics degrees, let's say.
Jason Jacobs: And where did the idea for Ultra come from and and start back as, as far as you want or as recent as you want up to you?
Jon Miller Schwartz: Sure. Yeah. I'm mean a little background on myself. Ultra is my third company I guess you can call me a serial entrepreneur, though that's never been my
Jason Jacobs: Cereal with an S, not with a C. Right.
Jon Miller Schwartz: exactly. Yeah I studied engineering in undergrad at Harvey Mud in Southern California. And I always was passionate and excited about building robots in high school.
I was on my first robotics team, if you know what that is. It's a nationwide kind of robotics competition and, in college I continued working on robots. At Harvey Mud we had an underwater robotics team, which is significantly harder than building robots on [00:06:00] land 'cause everything has to be watertight.
But, ton of fun. And anyway, ended up starting my first company right outta college. Not robotics though. We were actually writing software for 3D printers. This was back in 2012, 2013, right as the 3D printing market was exploding. And, started the company knew nothing about how to build a business or, really even how to build product.
But through that experience, learns the, I'd say the foundation of what I know today. And we ended up selling that to MakerBot, which you may remember was leading that that boom in the 3D printing space. So that brought us back to New York City. Spent some time there and then we started our second company, actually myself and my co-founders.
So Ultra's, the third company we founded together. Yeah, so we,
Jason Jacobs: guys are like an old, married couple of entrepreneurs.
Jon Miller Schwartz: that's literally what it feels like. But it's great, it's like there's no other team I'd rather be working with. And so our second company, it was called Voodoo Manufacturing, and we basically built a high volume 3D printing factory. We, we asked ourselves [00:07:00] what would happen if we took a lot of these printers and tied them together.
With software and with robotic automation to build a factory that could compete with overseas injection molding. But here in the United States, and really the goal was to build like AWS for manufacturing. So a software driven factory, a software driven process where you could just like easily scale up production of any part.
And, iterate on that part without, the kind of sunk costs of a mold, let's say, which is what most traditional manufacturing requires. And at Voodoo, yeah we built a high volume plastic part factory and there we experienced firsthand how hard it was to get robots to do.
Useful things reliably. And so we built robotic systems in our factory to automate, various like manual tasks that were highly repetitive. And, we did it and it was super cool when we, finished these robots because they would just give our factory a superpower essentially.
But if anything changed, literally, during the season changes when the temperature shifted [00:08:00] and the whole building would like warp slightly the system would stop working. And so that was how brittle we're talking about for. Just like generally robotics. And that experience, I'd say was really fundamental to to us and our desire to actually have had like more flexible robots that could easily be adapted to different tasks and locations in our factory.
And yeah, that experience plus just over the past few years, paying close attention to what's been happening in the AI space and in the robotic space basically made us feel like now is the most exciting moments when we can actually go out and build these amazing new types of robots that can really, in our view, enable, you know, factories and warehouses in the United States, unlike they've ever been enabled before through automation.
I.
Jason Jacobs: And the it sounds like. What's now I'm gonna say this as a statement, but that's my way of asking a question and making sure I understand it right. But it sounds like what's now becoming possible because of AI is more versatility. Where if [00:09:00] there's different tasks that come at it, or as the conditions change around it, it can sense what's happening or what's needed and adapt so that it doesn't start to, cha domino chain of effects where it's in order to change that this needs to change by this person, which then pulls them off of this task, which then means that this doesn't get done. Which, that it's no, it can just be smart and adapt in real time without screwing anything or set setting anything else off kilter.
Is that right?
Jon Miller Schwartz: Exactly. So I would actually say, so there are two things that kind of like this new method of AI based control allow for. So the first is what you're describing, which is essentially. Automating tasks that are unconstrained and even can occur in like open world environments. And so what I mean by that is, typically with a robot, if you wanna automate a task, you're going to constrain everything about that task and that environment.
That you can. So if you wanna pick up a [00:10:00] water bottle, you wanna make sure that water bottle arrives at the exact same location every single time. So the robot knows where to position itself to pick it up. And so you wanna just reduce as much variability in the process as possible because that, allows for the kind of highest level of reliability for your robot to work.
And the kind of beautiful thing about using neural nets for robotic control that are trained on data is they can adapt to that variability. And you don't need to cover every possible edge case when you're building your robot because it essentially can adapt, just like you as a human would adapt.
You can look at the object and reposition yourself slightly if you need to pick it up. So that's half of it. The other side of it is controlling robots that traditionally would've been very difficult or impossible to control. And so an example of this is a humanoid robot where, you have a robot that has many degrees of freedom, many joints, hands are incredibly complex.
And so the kind of traditional control methods, really wouldn't work if you wanted to be able to dictate how a [00:11:00] hand should move to pick up an object. They would be like, gross simplifications, but you couldn't get the human level dexterity that we have with our hands. But with this new method we can basically take data of robots doing these types of tasks.
And I can go into that more how you got that data. But you can take that data, train a neural net, and you can now start to control, complex robots like a humanoid. Or other robots that would've been impossible to control previously.
Jason Jacobs: Huh. And so when you saw these trends how did you do did you think about gosh, we could. Innovate here and then go find some places to point over time or did you have specific use cases in mind and markets to tackle and then go deep in understanding those markets and work backwards?
Like it seems like there's a bit of a chicken and egg thing in terms of well, can the technology do it? And that's a whole set of time and resource to figure that out and to make it able to do it. And then there's also but if we did it. [00:12:00] For whom? And would they pay and would they care? And would it matter?
And would the, what would the could we do it in a way where the economics work? And we can do it profitably over time. So how did you think about that chicken and egg in, in the case of ultra.
Jon Miller Schwartz: Yeah, there's this like joke in the robotics industry where, you know, just like people who wanna build robots are just excited by robots, but everyone is they have the hammer, right? Which is the robot, and you're looking for nails. That makes sense. And, result in like a healthy business and the industry is littered with like dead robotics companies.
There are many over the past couple decades that have tried and failed to build something that could address a wide enough, need. So I guess I'll start by saying our fundamental belief is that. There is a huge amount of just latent demand in the world for labor.
And we as a society I think, we want to be ambitious with the projects that we take on, but often it doesn't make financial sense. We can't find the money to go and, do these [00:13:00] projects. We can't organize the labor or it's too dangerous, or just people don't wanna, do these things.
And so I think that if we could. Create essentially economical, scalable labor that was capable of doing, a lot of the work that we as a society want to do. There's just gonna be a ton of demand for that. Like everything from building cities to what happens in factories and warehouses.
So the question is like, how do you get from where we are
Jason Jacobs: So does it, does that presume that that the laborers that build cities or work in factories, that those are not desirable jobs?
Jon Miller Schwartz: I think some of them are, my sense from talking to people in American warehouses and factories is there's a labor shortage. There are, basically high churn jobs where companies are constantly trying to find people to come in and fill them. Whether that's because people don't want these jobs or just literally, the population is not sufficient to fill these things.
I think it's maybe a little bit of a mix of the two, but if you go into many, modern factories and warehouses. The types of tasks that we have people [00:14:00] doing are being reduced to very robotic tasks. Any like Walmart or Amazon warehouse has probably constrained people to stand in the same, five foot radius and just do highly repetitive work, like packaging orders.
I don't think these are great jobs that, I don't think there are a lot of people out there who think to themselves, man, when I grow up I really wanna, do this super repetitive, boring thing. But I also just think that, again, it's like hard to organize and find labor for, all the projects that we want to take on because it, it's also a question of economics, right?
And, making, make, making these things happen for the price point that we, people can be paid at.
Jason Jacobs: And so how. How much time did you spend looking at potential use cases and applications before anchoring and declaring either externally or even to yourselves that this company was a worthy endeavor?
Jon Miller Schwartz: Yeah obviously we have a background in manufacturing. And we're not focused on manufacturing right now, but I think eventually it is an area that we wanna grow into. My [00:15:00] co-founder, max spent some time working. He was a consultant at PCG before we started Ultra. And so he spent some time working in the logistics space.
With USPS and Amazon Robotics, an auto store. And so we had some insight into that industry. And basically just one, that industry has grown a lot over the past. Not just decade, but even during the pandemic, right? There was a huge surge of online ordering. And so all of these warehouses rapidly needed to scale and try to find people to fill roles.
And in a way it's an industry that has been most primed to adopt robots. Like when you think about like Amazon and their use of, mobile robots that move around like racking with products in them even to robotic arms that are moving packages around. It's probably, they're most commonly seen in warehouses.
But that's really only true of the biggest players. And if you look at the rest of the industry, most warehouses have very little to, to no automation. And that just comes down to the cost and the challenge of integrating these systems and [00:16:00] making them run day in, day out. And so that's how we ultimately landed on Legi, the logistics space as where we wanted to focus.
And then. Specifically order packaging for e-commerce orders. And so our robots today are packaging up orders that you know are getting shipped to your home when you place an order online for products.
Jason Jacobs: And just from a timeline standpoint, when did the company start? And then what phase are you in today? And then it'd be great to work backwards too. It's okay, if that's the phase you're in today, and that's when the company started now fill in the whatever chapters we missed. Yeah.
Jon Miller Schwartz: Yeah. So we started the company, we incorporated about one year ago and when we got
Jason Jacobs: that's so still pretty recent.
Jon Miller Schwartz: Correct. Yeah. Still recent, incorporated a year ago. We really only started on this direction about nine months ago. We, when we started, it was actually from a place of getting the band back together.
I had just left my previous job. I've reached out to my co-founders, Oliver and Max, and I said, guys, we should start another company. I think it's time. And once [00:17:00] they were on board we all, agreed that we were gonna dive in head first.
That was about one year ago. At that time, we knew that we wanted to do something in the robotic space. We were still evaluating a number of different options and in some ways we were trying to avoid making the robots ourselves, from our previous companies.
We knew how hard hardware is. And at first we were looking at some software only ideas. We spent about three months talking to as many people in the industry as we could. Which, you always look back on this phase, I think when you're in it. It's frustrating 'cause you just wanna find the thing, right?
And just start sprinting forward. But in hindsight it was like incredibly important and informative and even just the connections with a number of the people we spoke to have proven invaluable. So that was the first, like three months of the company. And finally in about June of last year, we landed on our current direction and, yeah, we were just off to the races then trying to start developing the tech and building our robots.[00:18:00]
Jason Jacobs: Got it. And and what is it that led you to believe that building robots was necessary given that you came in hoping that wouldn't be the case?
Jon Miller Schwartz: I think there were two problems we found when we looked at the industry as a whole and the potential to, to sell into the industry. First off, like it's young, right? Robotics have been around for a long time, but this kind of like new. Model of a robotics company. It's really only been around for, two years most.
And there's just not a big market, or at least there wasn't a year ago of companies that, maybe we could like build like tooling or software for managing robots.
Jason Jacobs: so a lot happening in the lab, but not a lot happening in the field.
Jon Miller Schwartz: Correct. Now, that was a year ago, honestly, a year later now. It's been insane to watch. There's just been this like Cambrian explosion of robotics companies that every kind of layer of the stack, and that is what we assumed would happen. And it's, exciting to see that is happening. But [00:19:00] even still, compared to other industries it's still a small industry. A lot of the companies are, very young and they themselves are figuring out exactly what they're doing. And so I think it's tricky to try to sell into those types of companies. The other challenge that we found, and I don't know if this is just something about founders of robotics companies or, I guess I have some ideas here, it's hard to sell into robotics companies 'cause.
Most of them want to do everything themselves. I think this is changing. We're not actually doing this. We're using off the shelf hardware to build our robots. You go talk to many other robotics companies and they are. Doing everything in the tech stack. They're developing their own actuators.
They're assembling and manufacturing their own robots. They're building all of their software that they use in house to basically, control and manage their robots. And so we just kept getting, I. We just kept facing dead ends when trying to talk and sell to these companies.
And yeah, ultimately we wound up at a place where we just realized if we wanted to do [00:20:00] something in this space, we were gonna have to like, build the robots ourselves.
Jason Jacobs: And where are you today? How far have you gone?
Jon Miller Schwartz: We've gotten what I would say surprisingly far, I'm surprised myself, we spent the first few months of the company purely focused on the AI side of things. Given that was the most recent breakthrough and in some ways the least mature of the tech stack, we really wanted to convince ourselves that these new AI control methods we're.
We're actually capable of doing the things that we wanted to do with them. And so we spent a few months like purely focused on, on research and evaluation where we were using hardware and recording data with that hardware and training our own neural net models. And then just like making sure that do these actually work?
Can we actually get a robot to do a task reliably with this new paradigm for control And. The answer was yes we convinced ourselves and that we could do that. And yeah, we spent the first three months doing that. We actually went through Y Combinator over the [00:21:00] summer. We were out in San Francisco for three months.
I'm, we're based in New York. We live in New York now, but we moved everyone out to sf, including my family. So that was a whole sitcom type experience, but ton of fun. And yeah, so you know, we went through Y Combinator over the summer coming out of that. We raised the seed rounds in September and then we came back to New York and the focus shifted from kind of the AI side of things to let's actually go build and deploy these robots and validate the.
Customer demand for this type of product, validate that we can build a robot that can actually do this task in a way that is attractive in terms of the cost and the throughput. And yeah, since since the fall, we've deployed now multiple robots that are operating in our customer's warehouses and our packaging orders every single day.
It's like an insane timeline, in my opinion for developing and deploying robots. But I also think [00:22:00] it's like indicative of kind of the point we're making, which is existing hardware is good enough to build these types of robots and the majority now of the work to do and of the value to create is actually in software and just like deployment of these systems and getting them to work well for customers.
And so that's really where we're spending most of our time today.
Jason Jacobs: And what types of customers are you focused on? Who do the buyers tend to be within these organizations, and how do they tend to be doing these tasks today? I.
Jon Miller Schwartz: So we are currently selling into the logistics industry and two, three pls, third party logistics companies. These companies, they're the companies that make the American, logistics industry work, right? So it's they receive inventory often coming from overseas. They, store it and move it around between warehouses.
And then when you place an order online, someone or a robot sometimes in the warehouse goes and grabs that [00:23:00] product, puts it into a bin, that kind of has your information linked to it and. Eventually the end of the process a human today is standing at a workstation and takes those items that you've ordered online and puts them into a poly mailer or a bubble mailer or a box, and that gets shipped to you.
And so yeah, three pls are companies that, that do that and much more. Honestly, it's a huge industry. We're selling to those companies today. They vary in size. They're the, super large players including the DHLs and FedExs of the world, but it's incredibly fragmented industry, and so they're also a lot of.
Companies in the mid market or smaller ends of the market who are almost regional and serve, specific areas within the United States or have specialties in terms of the types of products that they they store and ship. And so those companies, in including many of the large players, this is entirely un automated.
And they have people today standing AT stations and packing up orders. And so we're currently selling them, [00:24:00] them robots to help, basically augment those people that they have doing this and to help them just scale as the industry grows.
Jason Jacobs: And what function and level of person do you find is the best entry point? So far and what's the pitch?
Jon Miller Schwartz: Yeah. It depends on the size of the company and the type of the company, how technology innovation forward they are. So far we've had a lot of luck, with smaller to medium companies going directly to the business owners or the kind of sea level team.
But as the company grows, you'll typically see a innovation or, automation kind of department. And so there are people who are constantly thinking about how do we identify and try out and eventually scale like new innovative technology solutions that can help us. And those teams are our go-to when you're talking to companies that have them.
But a lot of times they don't exist. And so you're either talking to, like I said. The business owner or kind of the head of operations, who's the one who needs to figure out, like how do we [00:25:00] actually get our packages out the door on time every single day, 365 days a year in some cases.
And so the pitch is really about helping them. Have reliable, affordable, scalable labor, right? As I mentioned before, it's not it's very common to talk to people who run these businesses and hear that they are having a hard time basically getting, workers in the door to fill the demand that they're seeing from their customers.
And so the benefit of a robot is, it's consistent throughput. So you have a robot or multiple robots in your warehouse and you know that they can do, x thousand packages a day. You can rely on that seven days a week. And that's really the main pitch there. Beyond, the kind of ROI that we're trying to hit and increase over time, which just comes down to, the cost of our robots and their operation, and then the throughput that they have.
So how many packages can we pack an hour?
Jason Jacobs: Uhhuh. And do you find, are they seeking augmentation to fill. Unfilled seeds are they thinking replacement as [00:26:00] it relates to humans? And I guess I'll ask this both in terms of their short term aspiration, but also their long term aspiration to, to, to the extent that they're thinking about and talking about that.
Jon Miller Schwartz: So we have not seen any of our customers. Replace any of their workers with our robots. This has at this point has really purely been an augmentation strategy. Our customers are all growing. Like I said, this industry has been growing. For a long time now, and I don't think that's gonna stop anytime soon.
Especially, we can talk about the tariffs more, the tariffs have also meant that like more fulfillment is now being done in the United States as opposed to, in our neighboring countries or overseas. And all of our customers and all the companies we talk to are trying to scale to meet this demand.
And really the way we see this carrying out is that, these robots are additional to the labor that these companies already have in some cases. I think it's [00:27:00] also about shifting labor to more like useful, like well-suited areas within the warehouse. And so for instance, one of our customers has bulk packaging operations.
And these are like products that they just ship, many of every single day. And so they would have people who are just like in the back of the warehouse, just like constantly packing the same exact thing over and over again. It's a task that no one wants to do, and they'd rather be doing things that feel like they can actually like.
Flex their, like human skills a little more. And so as we've taken on more of that bulk packaging, those people have been freed up to basically do other things within the warehouse. Like focused on optimizing the the inventory within the shelving and packing orders that are a little more complex and require more of the kind of human skill.
Jason Jacobs: Huh And just from a from an effectiveness standpoint. How does the robot compare to the human packer? And from a cost standpoint, how does the robot compare to the human packer? And then what do you think will happen directionally as as ultra and [00:28:00] as the category matures?
Jon Miller Schwartz: Right now in terms of performance, our robot is subhuman. We cannot pack as fast as people do. People are amazing. By the way, the human body is amazing. Our ability to move quickly, but without breaking things to respond to our environment through vision and what we hear and touch.
It's like the more you get into building robots, the more you just appreciate how amazing the human body is and how it's evolved. So yeah, with that said, we're not at human level in terms of like packaging speeds right now, but we're not that far away. We have our robots, we pair with other hardware that basically supports the packaging process.
And means that like we can get to like pretty high speed. Recently we hit about 200 packages an hour. Now, like when you think about speed, it also depends like what types of orders you're packaging. Are they orders that have a single item or multiple items or the items, relatively small and easy to grasp and pack?
Or are [00:29:00] they large and floppy and maybe you need to fold them to fit them into the bag. So all of that is going to affect your throughput. We obviously we've only been doing this now for nine months, and so we haven't spent a ton of time on the actual optimization of that process.
We see a lot of opportunity to go and improve the speeds here. And that's really gonna come through trial and error. So doing more of this means that we can collect more data, and the more data we have, the more we can optimize our neural net models to make them better and faster.
In terms of cost really we're trying to come in at about the same cost as a single shift worker. But then the idea is that you can run the robot for multiple shifts. And if it is equivalent for one shift, you can now take it and have it run instead of just eight hours a day.
You can have it run, for another half shift or another full shift. And so you get a lot of value out of it the more you run it. And our goal is to basically enable our customers to run. More than they were running previously. So get more utilization out of not just the robots that we're selling them, but like their entire [00:30:00] warehouse.
And at the end of the day, the challenge there always comes down to labor.
Jason Jacobs: And and given what you know, after so many years working in robotics and also just seeing how quickly things are evolving in the category. Over the last year, as you mentioned. If you were one of these human workers how would you be scared? Would you be excited? 'cause it's, and I asked because I mean I talked about this when Hadley Harris came on the show, it came up about jobs and stuff.
And it's look, on the one hand, there'll be a lot of shifting. And when there's shifting, there's change. And then some segments can suffer or need to re-skill. Or whatever, but then it can be curing diseases, it can be like, there's all these societal benefits that are unheralded. And you can't discount the benefit too.
You can't just look at the negative. But I'm, so I'm just try, I guess trying to level set, like how much change will be coming and what will the implications be for these workers with the understanding [00:31:00] that that whether there's change or not for these workers, that doesn't mean it's. It's necessarily bad or shouldn't happen.
Jon Miller Schwartz: Yeah, I the response so far from the teams that we're working with has been. Pretty overwhelmingly positive. And I think that's because for these teams, like they view this as a way to like boost, their productivity, not just in the individual productivity, but like the companies, the team's productivity.
If you walk into most of these warehouses. They're overwhelmed pretty much every single day. There's always the clock running and we need to have these orders packed by, whenever they get picked up in the afternoon and it's just like day in, day out. So they want whatever help they can get to take a little bit of that that burden off of them.
And like I said, I've now stood at these stations and packaged orders myself, and I don't view this as a, a great human opportunity. It is incredibly repetitive. It's exhausting. You're standing on your feet all day and your knees start to ache at the end of the [00:32:00] day. Do I think that like long term there may be some like.
Job loss in this specific space, probably. Just like there is across all, automation when we figure out a way to go and replace some highly repetitive process with something that's a little more economical. And yeah, at the end of the day, I think this is a larger question that's, obvi, obviously I need an opinion here, but, we all need to think about, which is how do we retrain people for the types of things that they are uniquely well suited to do.
Which by the way, I think there's an incredible amount of opportunity out there.
Jason Jacobs: And I guess I'll ask the, I'll ask a little bit of a different question than what is standing in the way of the robots doing? The this full function today, and how long do you think it will take for us to get there?
Jon Miller Schwartz: so just to give an idea like. I have I have two daughters. One is six months and one is three years old. And it's really interesting to watch kids develop, [00:33:00] in terms of their physical ability, right? Their ability to respond to the world, to reach for things, to eventually grab those things, to manipulate them, to bring them to their mouths or, control them and eventually to do things that are a little more complicated and dextrous.
My 3-year-old, it, it's exciting. She's now able to start playing. We have a card game and she can pick the cards off the top of the deck pretty well right now, but she doesn't get it every single time. And when I think about the progress we're seeing in robotics, like maybe a good way to think about it is to compare it to like the development of a child.
Cutting edge robots today in terms of like dexterity are still in the, six to 12 month age range where maybe they can grab items and you can reorient them a little bit to, to put them into a different location. But there's so much that we're not able to do both at the software level and at the hardware level.
Like even just having, robot hands or end effectors that have the the finesse to be able to, use small tools or reach [00:34:00] behind areas where they can't fit. Again, going back to what I was saying before, I think that there's just like going to be a ton of opportunity for people to learn how to do things that, robots won't be able to do for a long time.
But it may require some retraining and some, new skills that people don't have. And I think that's in many ways one of the things that we should be focusing on as a society is how do we do more of that?
Jason Jacobs: Uhhuh and and so I mean if you play out, call it the X twelve, eighteen, twenty four months for ultra. It seems like there's different ways you could go. You could just figure out how to scale the things you're already doing and serve a higher percentage within each facility, but also serve more facilities, more customers.
You could take on more tasks within the facilities than just the ones that you started with, or you could also expand to different kinds of facilities and different kinds of use cases over time. How are you thinking about staging and phasing, and what are your key priorities?
Jon Miller Schwartz: At this point we are. [00:35:00] Focused singularly on, on this one task. We believe, startups, you need to have a deep focus and do whatever you're doing really well. So we're in the near term, purely focused on order packaging for e-commerce fulfillment. Our our focus on our strategy is to basically deploy as many robots as possible because with more deployed robots, we are generating more data.
And like I said before, we believe that really the key to providing increasingly valuable and capable robots at this stage is going to be about the data that you can. Collect and generate and use to train your ai. In the same way that we've seen over the past couple years, LLMs, scale up and become more capable and now they can, do images and video and audio.
That same type of scaling is going to apply to robotics. And so again, the more robots we have out in the world. That are collecting, a variety of diverse data, grabbing different types of items in different warehouses with slightly different process, all of that's going to enable us to build the most [00:36:00] capable, packaging system.
And then at some point, a few years down the line, we'll take that and then we'll move it to the next area in the warehouse where we think we can add a lot of value. Of which, there are a number of other tasks that have a similar type structure where, there's a person standing in a single spot, moving items from one location to another.
And so we wanna position ourselves to slowly increase to like those types of tasks over time.
Jason Jacobs: How do you think about competition and defensibility?
Jon Miller Schwartz: Yeah. At the end of the day, I think one this this industry, this opportunity is so incredibly large. It's not gonna be winner take all, at least for a long time. And hopefully never, right? Because even the idea of winner take all in this space is scary. Otherwise when we think about like defensibility, it's really about.
Owning the relationship with these customers and then having robots that are operating in their warehouse, like I said, generating data. I think that there is a large challenge to taking a [00:37:00] robot and putting it in a customer's facility and getting it to work day in, day out, providing positive ROI.
And once we have a robot in the warehouse, that also then, increases the chance that customer's going to wanna work with us for their next robot and the next one. And so I think, our product, our service is very sticky from that perspective. And so yeah, at this point, we are trying to, one, just grow into our current customers, but then also go and start building relationships with other companies that we see a lot of future opportunity with.
And as we have more robots, that kind of then grows the size and the rate of the flywheel, more robots creates more data. More data, makes our robots better. And our robots being better means that we can sell more robots.
Jason Jacobs: As I've been investigating ai, it seems like again and again what I'm hearing is, launch early and then turn your customers into source of data that can train the model. And that way you're training the model over time based on real customer data versus just [00:38:00] stuff you can grab here and there from random places and it'll get you more bulk and also just better.
Data. When it comes to training the model or training the neural net as you said, is it the same skillset that does that across data types and data sources and across industries, et cetera? Or is it wildly different and what is that skillset?
Jon Miller Schwartz: Do you mean So, so different from training, let's say like a, a language model
Jason Jacobs: Yeah. Or like I had so Swept Mill the founder of a company called Swing Vision using AI for racket sports. And he was talking about how yeah, they grabbed some data of clips of people swinging stuff, but that once they were in market, then the data was flooding in and it was from all the people that were using the app.
And so that is, I would imagine. A very different kind of data and model training than robotics for e-commerce fulfillment, but that's what I'm asking. Those are just two, but I'm sure we could just spitball and talk about dozens of [00:39:00] different use cases in industries and data types.
But is it the same sport in terms of training the model? Or does that sport vary widely?
Jon Miller Schwartz: generally it, there are a lot of similarities and someone who is training a model that a model to help people improve their racket sport game, or predict, where the ball might be going. A lot of similarities between that and what we're