Episode 102 – The Future Of Food Automation: How Chef Robotics Tackles Labor Shortages With Ai Robots

Food Robotics Automation: In this episode Bill Brock interviews Rajat Bhageria, CEO of Chef Robotics. They explore a groundbreaking approach to addressing labor shortages in the food industry through AI and robotic technology. They talk about the challenges of food manufacturing, the intricate process of developing flexible robotic systems, and the innovative “Robotics as a Service” model that’s transforming food production.

Rajat explains how Chef Robotics targets high-volume food manufacturing facilities. They focus on food assembly tasks that represent 60-70% of kitchen labor. The company’s unique approach involves using six-degree freedom collaborative robots with configurable hardware and AI-powered software that can adapt to various ingredients, tray types, and portion sizes.

The episode highlights the critical labor shortage in the food industry and how Chef Robotics’ solution goes beyond traditional automation. By collecting extensive production data and continuously improving their AI models, they’re creating a scalable solution that will revolutionize food preparation across different sectors.

About the Show

“Over the Air” is a podcast that features unfiltered conversations with executives about their IoT journeys. The mistakes they made, the lessons they learned, and what they wished they’d known when they started. In each episode, our expert guests make bold predictions, reveal unpopular opinions, and share stories you won’t hear anywhere else.

Buyers Guide – How to Choose the Right IoT Company: Very

Subscribe here

YouTube

Apple Podcasts 

Spotify 

Transcript

Hey, everybody. Welcome to the Over the Air podcast.

Today, I have a fantastic episode for you all. We have Rajat, CEO of Chef Robotics. Welcome to the show.

Yeah, thank you. Thank you, Bill, for having me on. Yeah.

So, over the last few months, we’ve been focusing very heavily on talking to advanced robotics companies, and I came across Chef and was immediately blown away by what you guys were doing. So, would you mind just giving us like the 30-second intro, maybe elevator pitch? Yeah. So, at Chef, we make AI-enabled robots with food industry.

The food industry is actually one of the biggest labor forces in the US, and the big pain point right now is that it’s really hard to hire in the food industry. It’s actually one of the biggest food labor shortages in the US. So, what we essentially make is AI-enabled robots that can essentially help these food companies overcome that labor shortage and increase production volume.

And so, are you focusing and like most people have worked in a restaurant at some point, so do you focus more on like a fryer or like a prep kitchen, or what is like the robotic setup actually looks like? It’s a good question. So, we’re actually starting in much higher volume facilities, specifically food manufacturing, at least for today. The idea there basically is that it’s actually much more amenable to robotics today.

We can talk about why that is. So, we’re starting there as a kind of our step one, and the idea is that if we can and when we can kind of really deploy lots of robots there, we’re collecting a lot of training data, and over time, we can use those deployments to then kind of do more restaurants and fast casuals. So, food manufacturing, explain like one click into that.

It’s not like food processing, you’re not packaging chicken. What are you doing? Yeah, exactly. So, basically there’s two kinds of manufacturing, broadly speaking.

There’s low mix manufacturing and there’s high mix manufacturing. Low mix basically means you have a fairly small number of SKUs or products, right? You know, I’ll sort of like, let’s say the food use case. Let’s say I’m Apple, I have like 10 products, whatever it is, I’m going to set up dedicated custom manufacturing lines for each of the products, and they’re just going to do that product all day long, every single day, right? That’s low mix, and food and equivalent might be like Campbell’s Soup Company.

Campbell’s, you know, has a dedicated line for their chicken noodle soup, and they just do chicken noodle soup all day long, and they have a different line for vegetable soup, and they just do that all day long. That works if you have low SKU mix. In other words, not very many SKUs.

If you, on the other hand, have 200 SKUs, it’s very hard to set up dedicated custom lines for each of the SKUs. So instead of what you do is you set up kind of flexible lines. An example of this might be, let’s say you’re eating a salad.

Everyone wants a different kind of salad, right? So these companies have dozens of different SKUs, and they’ll run an SKU for two hours, and then they’ll change over the line, and they’re going to do a different SKU, and they’ll change over the line, and they’ll do a different SKU. So essentially, the way you can think about it is like most frozen meals, most fresh salads, burritos, sandwiches, wraps, you might find at a grocery store, airport, airplane food, K-12 meals, meals you might find in a hospital system. They’re actually made in a central facility, essentially a food manufacturing plant, where you have essentially a long and rare line, and people scooping food from big tubs into individual trays, and they’re doing it all day long.

So that’s kind of where we come in and help them out. That sounds like a very unglamorous job for the people in the loop there. Just scooping, scooping beans for burritos for 10,000 burritos in a day.

That would really be very boring. Yes, and it’s also very hard physically. I mean, just to kind of give you some context, it’s usually cold in these rooms.

It’s refrigerated, so it’s usually around 34 degrees Fahrenheit. And on top of that, you’re also kind of doing all these redundant motions, right? So you’re literally like your box aching and your arms aching. It’s just a very painful tub, I would say.

And how skilled are the robots? Obviously, you’re the CEO, you’re incredibly skilled. I took a lot of pride. At one point, I worked in a deli, and I was like, I’ve got the perfect sandwich wrap.

I can wrap the deli sandwich perfectly. So how fine grain are the robots? It’s a good question. So basically, the way you can think about it is, essentially, to make food, there’s three big tasks you have to do.

You have to cut food, you have to cook food, and then you have to assemble food. We really focus on the food assembly part of it. And the reason we focus on food assembly is because food assembly, from a labor perspective, is actually 60% to 70% of the labor that goes into a kitchen, basically.

Now, by the way, fine dining is a little bit different. Fine dining and cooking is actually usually the main source of labor. But for most non-fine dining applications, it’s really assembly.

So we focus on basically assembling a meal. And then within assembling a meal, we cut out certain things that we cannot do. So for example, we do assemble burritos, but we won’t wrap the burrito, or wrap, right? We can put toppings onto a pizza, but we won’t put the dough out or something.

You get the idea. Really kind of like taking pre-cut and pre-cooked ingredients, scooping them, and then placing them into the right part of a tray, a sandwich, a wrap, whatever it might be. Now, even that is still quite complex.

And why is it complex? First of all, there’s thousands of different ingredients. And by the way, if you cut the ingredients a little bit differently, or cook the ingredients a little bit differently, it is different. And so your robot has to be able to sense that and react differently.

I’ll give you a very practical example. If you think about certain ingredients like grits or something, sometimes it can be very unsticky, sometimes it’s very sticky, and your robot has to be able to very dynamically figure out what to do and be consistent and not spill the food, things like that. Every customer has different trays.

Sometimes they’re round, sometimes they’re compartmented, sometimes they’re triangular, sometimes they’re whatever it might be, right? So you have to be able to react to that. There’s different portion sizes. There’s different placements within the tray.

There’s different conveyors. So anyways, all I mean to say is there’s all these different variations of that, right? And the dexterity of the robot then is being able to say, okay, well, how do I be able to kind of flexibly automate that process for a new customer without having to do a bunch of custom software or custom hardware, right? In other words, you could imagine solving what I just said and saying, okay, we’re going to make a machine, basically we call it a machine that can put this kind of bell peppers that are cut this exact way into compartment one of this tray, and that is it. And you could put some custom software and hardware that could do that, right? But that’s not of course scalable.

So our, what we do at Chef is build a product which essentially have hardware that’s configurable and have software that’s configurable using AI models that essentially can do any of those configurations without having to custom stuff. That’s, that’s what’s hard, I guess. And what was the like early conception of this product versus where you are today? So you’re, you’re crazy entrepreneur, you’ve got an engineering background, you’re like, I’m going to go solve manufacturing and food maybe, or I’m going to create a robot that cooks.

Like where did you start? Yeah, it’s a very good question. I mean, I think like where I started was, okay, well, I believe that the biggest impact of AI, we were talking about that kind of the before we started here, like, I think the biggest impact of AI is going to be in robotics, right? And so I was like, think about, okay, where do I start? And I think food was actually very attractive, because it is actually one of the biggest labor forces in the US. The only labor forces are actually bigger is salespeople and nurses, nursing aides, I guess I should say, those are both categories that didn’t think are doable by AI and robotics today.

So food actually, for me, was like the biggest market I could possibly go after. And then, you know, talking to customers, they all basically said focus on assembly. So, so all of this was actually the same, like food was the same, the problem being labor shortage was the same, focus on assembly was the same.

The big thing that was actually different from, from the initial conception was, initially, we were focused on fast casuals and restaurants. That was our actual kind of thought about where to focus on. And by the way, at the time, there’s a lot of tailwinds for why this could make sense.

Chavs Kaunik had just kind of started Cloud Kitchens. Ghost Kitchen. And right, Ghost Kitchens were really starting to kind of blossom, right? A lot of investment into the space.

Obviously, this is pre we started in 2019 pre COVID. But even then, 40% of volume for restaurant kitchens was delivery. So what all of them were essentially doing is opening a back of house lines.

So you imagine going to Chipotle, there’s a front of house line, there’s actually a back of house lines just for delivery orders. And more and more brands were doing that. So that plus the proliferation of Ghost Kitchens made us feel pretty excited about this idea of what if we have kind of robot modules, physically quite small modules, that you could put into the back of house so which port layers are fast casual, or Ghost Kitchen to make your delivery orders, still assembling the tray, but for, for restaurants, and Ghost Kitchens.

That was initial conception. The reason actually we moved away from this is fairly simple, actually, which is, if you think about a fast casual or Ghost Kitchen, they’re not doing a ton of volume, they’re doing, let’s say 1000 trays per day. So what does that mean? Anyone who works there, any person who works there is a generalist from a labor perspective, they’re doing every single ingredient on the line.

It’s not like in a manufacturing of stations, right? Bob is doing the leaf greens, and Joe is doing the the sauce. And you see this in car manufacturing, and really any kind of manufacturing too, right? Like before, when car manufacturing was low volume, you had one, you know, 10 people kind of bundled around the car, and everyone’s kind of working on that one car. And then, of course, Henry Ford’s invention was the assembly line.

He said, Bob’s gonna do this, and Sally’s gonna do this, and Jane’s gonna do this, and etc. So that same thing is true in food. If you have low volume, everyone’s a generalist.

So Bob who works there is doing 90 ingredients. And so if I want to have a one-to-one equivalent to Bob using a robot, I also need to do 90 ingredients. If I cannot do 90 ingredients, then Bob is not freed up to do something else.

So what is that sweet spot for you guys, the number of ingredients then? Right. So basically, it has to be quite a small number. There’s not like in this number.

By the way, this is kind of going back in history, right? This is like what it was initially. Yeah, sure. At the time, like we were like, and by the way, the crux here, the thing that’s hard is like, why is 90 ingredients hard? The reason it’s hard is there’s no data set.

I can just go to the internet and download of how to manipulate food, right? So if you think about like, why has Chachi Pity and LLM has kind of really taken the world by storm? Well, arguably, yes, it’s a transformer. That innovation is important, and then attention mechanism. All these things are important.

But arguably, the bigger reason is there’s this treasure trove of data from the open internet, like Chachi Pity and GPTs are just trained on the open internet. That does not exist for robot manipulation data. There’s nobody that has built robot manipulation data sets around how to manipulate leafy greens or how to manipulate mashed potatoes that doesn’t exist.

The leafy green manipulation data set. Yeah, you don’t say. Yeah, there’s not.

It doesn’t exist. So that’s the trouble. That’s why manipulating 90 ingredients is tough.

So what’s exciting about manufacturing versus flash casual in ghost kitchens is that it is much higher volume, you’re doing not 1000 meals a day, you’re doing 10,000 a day. So what does that mean? Now, everyone who works there is a specialist, right? Like again, Bob is doing station one and station one is rice, for this meal, at least. Now, obviously, it’s high mix of these lines change over, like I said.

So throughout the day, Bob actually might be doing 10 ingredients, but 10 is a far cry from 90. Yeah. So initially, basically, what we did is we said, okay, look, we’re going to bootstrap a data set for like five to 10 ingredients, which is good enough to get some a single robot into production.

If you get a single robot into production, we’re collecting production training data. We’re learning how to really manipulate ingredients in the real world is somehow humans are cutting and cooking those videos on a daily basis. Now, using our production data, we continually train our models and they continue getting better.

And over time, as the models get better, the robot performance gets better, which means the customers continue buying more robots. And also means when we go to our new customers, they can already do the ingredients. So that’s essentially why we moved away from fast casuals.

But what’s interesting is like our vision is really to take this training data from manufacturing and one day also go back to fast casual skills kitchens, because we can now do thousands of ingredients, the broader data set, nobody else will have this data set, and you’ll be, you know, the only the only stop the one stop shop, or the only the only game in town. That’s the word. That’s very cool.

So like, where are you guys at with relation to deployments? Like you’ve got one robot in the field, 50 robots in the field, what’s what can you share? I know those things are always in the end to some some extent. Yeah, I mean, we have dozens of robots call over North America right now. So US and Canada right now, we’re actually expecting to expand into UK early next year as well.

They are like, I mean, you know, one metric I can share is we’ve now done 32 million servings across our fleet of robots, which is actually more than all the other food robotics. And these actually combined for all we know, at least. So you know, we’ve now done, I think, I think it’s 1700 plus ingredients of all disparate ingredients, I should say, right.

So like a ton of different varieties of ingredients that like, allow this data flywheel that I mentioned earlier to really blossom and be useful for our customers. What has been some of the like hardest learning? So like, what have been some of the larger hurdles that you’ve had to tackle? Good question. I mean, I think there’s a ton.

And there’s like, basically, for any dimension, you can imagine there’s a ton of hurdles. You know, I think probably one of the bigger hurdles that’s kind of top of mind for us right now is a lot of times we talk to customers. And they’re basically like, we didn’t even know this is possible, if that makes sense.

So a lot of the sales conversation is like, essentially market education, if that makes sense. It’s not like, hey, let’s have a conversation around like, you know, ROI and all that. Yes, we do that, of course, but a good portion of the sale is basically like, prove to us that you can even do this, prove to us it’s even possible.

We are used to dispensers and gravity fed kind of the traditional automation that’s existed for 40 or 50 years. And so I think, you know, one of the things that we are kind of really thinking actively about is like, just showing the world broadly that this is even possible to get people thinking about it. So that when we get in touch from a sales rep from our side gets in touch, like we’re right there, right? They already know what we do, or that’s possible.

And we kind of get right into it. I mean, that’s one learning, but there’s many others we can talk about. Yeah, on that, like looking at the robots, like if for those listening, if you don’t have it up, it’s really fascinating.

It’s kind of not what I would have thought of as far like the food prep robot, you think about the robot arm and the fryer, or maybe like the humanoid that’s like back there chopping it all up. And it seems to be a fairly different form factor. So was there anything around like the layout and like conveyor? Like, how did you guys come to this, this current level of design? Yeah, so I mean, I think the, the design principles we had was, was, was as follows.

We want to keep the hardware as generic as possible. While essentially over speccing it, and I’ll kind of explain what I mean here. So we want to keep it as general as possible.

We want to make the hardware over specced. And we want to move as much of the complexity of software as we can. So okay, what does that lead to? Well, to make it general, we do pick and place, we don’t do dispensing, right? So the system’s a pick and place based system.

So it has this, you know, robot arm basically. We did try a lot of different configurations. By the way, we tried lots of dispensers, we tried less degree of freedom robot arms, like three degree freedom robot arms, like Cartesian robots and, and, and such.

Finally, we kind of settled on a six degree freedom collaborative robot. And that’s basically because of the second point, which is over speccing. One thing about Chef is that we do robotics as a service.

So we’re constantly kind of helping these customers kind of onboard ingredients. And, and it really is a service, right? So we get into the door and we’re doing some ingredients, but down the line, the customer might want us to do something completely different. And of course, you know, as we go to our next customer, they want might want us to do something completely different.

So the way we think about it from a hardware perspective is like, how do we have a system that can do basically everything, even if we don’t need to do it basically everything today? Yeah. So six degree freedom, basically allows you to scoop, allows you to basically anything. For certain ingredients, you actually might not need that.

For like rice, you don’t need six doff, three doff is probably sufficient. But maybe down the line, you want to be able to scoop long pieces of asparagus. And it’s useful to have six doff in the, in the hardware, in the system, we call it modules, so that we can essentially mass manufacture the module and it’s exactly a module at every single customer.

It’s not like we’re making custom hardware, it’s like copy paste from the hardware perspective, we make thousands of these basically. And then the AI and the software really allows us to be flexible, it allows us to deal with the complexity I mentioned earlier, different ingredients, different portion sizes, different compartments, different placements within the tray, different trays, conveyors. And there’s been a lot that’s changed on a software stack in the last six years since you broke around on this thing.

What have been some of the advancements that have really helped you guys out? Yeah, so I think it’s Yeah, exactly. There’s been a lot of advancements. You know, robotics before essentially follow this idea like a, I think it’s kind of like a see, think, act model, right? So you have kind of a perception stack, which is kind of understanding, it’s taking image data from the world or video data from the world or live data, whatever kind of sensor data you might might want.

Then you make a decision, right, that’s where machine learning might come in. And then you act, which is like actuation, you pick something up or you turn a steering wheel or something. So this is how it kind of how robotics was kind of when we started chef and arguably, almost not almost all robotics that’s in production more or less today is that still what is becoming more popular nowadays, of course, with the advent of the mechanisms is is imitation learning, I would say.

The idea of imitation learning and then arguably Tesla FSDV 12 is kind of the first kind of odd scale production example of this. So essentially, what what Tesla used to do is that same idea, right? See, think, act. Now, they’re FSDV 12, for what I understand, at least, kind of uses imitation learning, which is to say, they say, let’s take the best x drivers in a particular city or area or the world, and let’s get training data from them, which is to say image data and the resulting steering wheel and brake and accelerator actuators, actuation data.

And then we can kind of have an end to end model, right, that the maps from images from their vehicle to to steering wheel angles, basically. And so it’s kind of an end to end model, right? And it’s kind of the human is doing demonstrations, it’s, it’s learning from demonstrations or imitation learning kind of same idea. And that has kind of been very exciting for the community, I would say, because you don’t need to write a bunch of code, you don’t need to have like, hey, when you detect a red light stop, or, you know, when you want to do a lane change, see if there’s space and then move.

I mean, I’m simplifying, but you don’t need to have these rules. But instead, you have more of an end to end system that the more data you give that the more intelligent guess. So anyways, that’s kind of also exciting for us.

So what we’re doing is we’re leveraging imitation learning using a technique called diffusion policy. And that basically allows us to demonstrate to the robot new tasks. And for us, new tasks might be new classes of ingredients.

And what I mean by this, by the way, is like, you know, if you if you if you want shaft to go from like diced chicken to diced turkey, then diffusion policy or imitation learning is not very useful. We already have a starting AI policy on how to manipulate diced chicken going to diced turkey is relatively straightforward. But if you want to teach it a brand new task, let’s say you want it to go from being able to do you know, long string beans to not not long string beans to being able to do it.

Well, diffusion policy and imitation learning is very helpful because now for an engineer, instead of having to write code for a week and then test it for two weeks, they can just do a bunch of demonstrations of how to pick up literally just suiting up and recording themselves. And wow. Yeah, exactly.

And, you know, we’ve come up with like an interesting mechanism to do it. Essentially, we, we have two side by side robots, exact same hardware. And on one of the robot, the human is actually driving the robot using their hand to kind of do the demonstration, the second robot’s actually doing the data collection, and it’s kind of on an occluded base, an occluded data from the second robot.

But you kind of solve this problem in robotics called the grounding problem. Like what I mean by that is my hand can do these weird motions like this, but that doesn’t mean my robot can do it. By doing two side by side robots, when the engineer does this motion with robot one, you know that robot two can do exactly, exactly the same motion.

So anyway, engineers kind of like translating in their mind the best way to go about it. And right, they can write true. That’s cool.

And you know that the motion is like possible, and it’s like doable for the robot, right? So so that’s the technique we use. And we, we do 30 to 40 demonstrations. And that allows us to kind of have a starting policy, we call a starting policy, basically, and how to manipulate that new task or in that new ingredient class.

But what we have learned, by the way, I will say is that, you know, that’s a great way to get a starting policy. It is not necessarily production grade, because it is just in the lab environment. So the way we think about this is like, the way to really make, you know, leverage these new techniques in AI robotics is basically take production data from the field, right, real production data, and seeing how the ingredients change day by day, and seeing how the randomness in the field affects the robot, take that data, plus the imitation learning data and kind of use it.

And that’s kind of how you get a really robust system that can withstand the randomness that you see in the physical world. So how long does it take for you guys to go from green beans to spaghetti squash, on introducing a new food? Yeah, I mean, to go from like one, to go within an ingredient class, it’s quite straightforward, it’s like five to 10 minutes, and we’ve got a lot of tooling to make that very quick, right? So in other words, like, maybe I take a different example than yours, and then I’ll answer your actual question, to go from like, let’s say we can do green beans, we can do green beans, and now we’re going to do asparagus. That’s like five to 10 minutes.

Okay. If we want to go from doing like, green beans to not doing spaghetti, completely new class, that’s where this imitation learning technology can help. Now, essentially, like to do the demonstrations itself, it’s like 40 to 50 demonstrations, which is a couple hours, that’ll get you a starting policy.

And then, you know, then it’s like production data. So now let’s take that to the field, let’s try it out, let’s test it. And that’s the part, actually, we’re kind of still perfecting, and that takes a little bit longer, a month or two, basically, to really get enough production data to like, really make it good enough to like, extend the test of time.

I’m just like pounding you with questions. Usually, I have more to say than this, but I’m just so interested that I like, can’t stop. You talked a little bit, and we’ve interviewed a lot of robotics companies, and I’ve heard of robotics as a service, but it’s not always a central theme.

It sounds like from from what you’re doing, maybe it’s just a little bit newer, or such a collaborative process, like, that the service aspect of this feels very obvious. And so can you like, talk to me a little bit more about what robotics as a service means to you, what it means to the business, how does it like impact even like the economics of the arrangement between you and the customer? Yeah, for sure. I think, you know, it’s a new model, right? So, you know, I think it’s actually very exciting that this is where the community is going.

Essentially, the way you can think about robotics as a service is, right now, if I’m a customer, right, let’s say I’m like, in our case, a food company, and I’m buying automation, then, you know, usually it’s quite expensive. It’s like, millions of dollars, half a million dollars that you have to put up. And obviously, that’s, I could never imagine somebody running a ghost kitchen, being anywhere near like, I want to do, like, pastries for my little pet project, no way am I going to bring in a robot that’d be way too expensive.

Yeah, exactly. So what’s cool about robotics as a service is basically, we say, look, you’re going to pay less than you’re paying your people, right? And for that, you get a kind of a system that can kind of end-to-end work. We’re going to provide the hardware, the software, the support, success, anytime something breaks, we’ll be there, like, it’s kind of an end-to-end service that’s less than your status quo cost.

So it really unlocks the market arguably. You know, and that’s actually quite exciting for a lot of people. What I will say is there’s certain, you know, for the very big enterprise customers, it is something new, right? They’re used to, they are used, they have plenty of debt, access to capital, they’re used to being able to buy equipment, and they have their business model figured out to be able to accommodate them.

So for them, it’s more of an education that we have to go through. Specifically, kind of, the reality of the situation is that a chef is a software product, right? So there is recurring cost on our site. It’s not like it’s a flat, it’s not like it’s a kind of only an upfront thing.

So we kind of have to kind of help educate them that, look, like, this is a little bit different than the traditional automation you’re used to. This is why the recurring model is necessary and kind of walk them through that. So it’s a little bit different between, like, newer upstarts or, like, you know, mid-market SMB businesses versus, like, the enterprise businesses.

And how amenable are they to software updates, maintenance? I know in auto manufacturing and some of these, like, lights out environments in China, I mean, I don’t know if you can lights out food manufacturing, who knows, but, like, even getting patches in place, software updates, can be really, really contentious in a manufacturing environment. Is that the same in food or not? Yeah, I mean, I think, like, I think it basically is, like, you can kind of partition it by, like, is the thing working or is it not working? The thing is working, don’t touch it. It’s generally kind of the philosophy.

And that’s true also in food, right? You know, the truth is that from a customer’s perspective, like, if they’re getting the output they need, they’re getting the throughput they need, they’re getting the quality they need, and there’s a bunch of metrics that matter, right? Like, for chef, consistency matters, spillage matters. If the metrics are good and they’re happy with it, don’t touch it. Like, even if you think you could do something better, there’s some risk associated with it.

So they’re like, don’t touch it. Or give me, like, an update every quarter or something. On the other hand, like, you know, there’s other customers where they’re like, look, they constantly get better.

They want to get more throughput. They want to get more consistency. They want to get whatever.

Maybe there’s some bugs. They want to fix the bugs. Maybe there’s, like, certain things they want.

And in those cases, they’re like, yeah, like, daily. Like, no, they don’t want you to test those in production. They’re not, like, you know, just ship me them and, like, test them in production.

But what they do is say, look, like, we do want the software update. We’re going to allocate an hour or two hours, whatever you need, every single day or every so often to kind of be able to kind of test your software, but then we expect it to perform well, basically. Yeah.

And then what we do on our side, by the way, is, like, you know, a lot of test infrastructure to, like, you know, obviously, you know, catch any issues beforehand. We’ll kind of do a lot of release tests after release tests in the office. And, you know, before it goes in production, we’ll kind of really harden it, right? And test it in the field as well, just to catch any issues that might happen.

We touched on this very lightly before the interview. But how did you even find yourself in this place? Yeah. I mean, I think, like, I kind of alluded to a little bit of it before, but I was very excited about the prospect of doing AI in the fiscal world.

I think that’s, like, probably one of the biggest opportunities that exist today. And then I kind of did a full broad search about, like, what product, company, customer segment should I go after? Food really made sense because it’s one of the biggest, I think. And there’s a very big kind of hair on fire pain point.

And then, you know, then I was kind of talking to customers and they kind of pointed me in the direction of assembly, food assembly, because it’s the biggest labor sort of labor pain point. And then they kind of put in, and then, of course, we learned about manufacturing and thought that’s the best way to market. So, yeah, it’s interesting.

It’s a bit of a, you know, nonlinear road, but I think, like, I think it makes a lot of sense what we’re doing. I think what’s cool is that so long as we succeed here, this market is giant in itself, but also we can take the same quartet to do a lot more. And what’s your, what are your thoughts on the current state of the humanoid robotics revolution? That’s kind of the obligatory robotics question, but like, what’s your, it seems like you’re probably, I would assume you’re bullish on humanoids.

Is that the case? Yeah, I mean, here’s what’s interesting, right? So I think, like, I think the technology is exciting. The learning, I think the learning from software in cars, I think I compared the software in cars. It’s probably the best comparison.

I think learning from software in cars is it’s going to take a lot longer than people might think, right? So like, if you take the time from when the drop or grand challenge happened, which is people might contest this, but I think arguably when the software and revolution really kind of started to at least kind of accelerate a little bit. And then Google kind of started their own project. But that was like 20 years to basically now having way more on the on the road and in some amount of volume.

So I think arguably general purpose manipulation is an even harder problem. You could, you can make that argument, or it’s a similar complexity, I guess I could say. So I think like, yes, it’s exciting, but I think the time horizon is a lot longer than people might think.

I don’t think it’s going to be five or 10 years. I think it’s going to be more like 10 to 20 years. And, and so, and so like, it’s basically a question of time horizon, right? I think what’s likely going to happen is that many of the new upstarts, they raise these massive rounds.

And that’s great to be able to kind of like get started. But then like, for your next round, those crazy evaluations, you need to have revenue, you need to have something meaningful. Yeah, you need to have a real business.

But like, these are like just core technology. Like they’re building their own hardware, they’re building new robot arms. It’s a very fundamental science and engineering, basically, as well as relying on these new AI techniques, both with like the hardware and the software, both hard, right? Both hard as in novel.

And so I think they’ll find it’s very complex and very hard to kind of not only get to production, which is hard, extremely hard in and of itself, but then get to scale, have enough progress and revenue, basically, to be able to raise the restaurant next round. And I think what you’re going to find is that many of these companies do fail. And there’s going to be a couple that kind of survive.

My intuition is it’s going to be the companies that are backed by an even bigger parent company like Tesla has, of course, all the money from the vehicle business. And then, you know, companies like Boston Dynamics, Hyundai, you know, any anyone that’s kind of backed by a big parent company, I think will succeed. Others, I think over the 10 to 20 year horizon will not succeed.

That’s just my prediction assumption. I think the technology makes sense. It’s just a very long time horizon.

Yeah, I watched a I got online this morning, there was Elon, he’s playing video games, talking about AGI being here at the end of 26 and then 10 billion optimist robots in the world. He’s notorious for making these claims on timelines. I do think that if anybody’s going to do it, he has the training set, he has mechanical capabilities and Tesla’s to be taken very seriously.

I think like, you know, but he’s usually it usually is a lot longer than he originally says. So I think both are true. I think it’ll happen.

And I think it’ll take a lot longer than maybe people think. Yeah, makes sense. What’s top of mind for you? What’s next for chef? I think I think, you know, like, I, we are quite excited about these modern kind of advancements in AI, which you call it kind of called out.

So I think kind of really investing in those efforts. It’s kind of a big paradigm shift happening, right from this, see, think, act model to more of an end-to-end system. And so we’re quite excited to use those technologies and accelerate our own work.

So I think that’s quite exciting. And then I think what’s also exciting is like, you know, we, we’ve now kind of released what we do. We were in stealth for quite a bit of time.

And now we’ve kind of released what we do. And, you know, I think the reason we did that is because like, how long in stealth? In three to four years. And mainly because maybe the reason was like, nobody was really thinking about food manufacturing as a pain point.

You know, most people assume that food manufacturing is kind of a solved problem. And we feel like we have the secret. And we’re like, let’s just kind of win there.

And the time is the secret. Right. Well, and it’s, you know, you have the exact right question.

Most people, when they think about food robotics, they think about restaurants. That’s like where everyone’s mind goes. That’s also where my mind went, by the way.

But then, of course, we realized this better market. And like, you know, we felt like that was really awesome. We want to focus on it.

And so now, basically, like, we feel like, you know, the product is really at a place where it’s ready to scale. You know, we have a lot of good customers. We have case studies.

You know, those current customers are buying more robots. And, you know, we’re really ready to kind of scale the go to market team and accelerate, like, and really kind of get to that goal of how do we how do we scale the robot shaft to thousands and thousands of robots all over North America, but also the world. So, well, this has been an awesome episode.

Thank you so much for making the time. It was a pleasure to meet you. What can we do? Is there any way we can help? Are you hiring? What are ways that we can increase your reach? Definitely looking for great people.

I mean, basically, in any field you can imagine, right, whether that’s like finance or manufacturing or supply chain or AI or like really anything that’s really tangentially related to robotics, you probably need somebody. So, I mean, I think that’s definitely top of mind. And, of course, if you’re a food manufacturer looking to automate, let us know.

I think more generally, you know, I would just say, I think there’s a lot of really smart people kind of getting into AI, which because of chat GPT and like that revolution that’s kind of unfolding in front of us. And I think it’s very exciting. And I would just encourage people to think about like, is AI going to be the bigger revolution or is AI applied to the physical world going to be the bigger revolution? And I think likely you’ll have the realization.

It’s maybe I’m wrong here, but I think it’s going to be the latter. Right. Yeah.

90% of GDP is in the physical world. The biggest market on planet earth is the labor market, right? It’s $45 trillion of GDP. So, so, so I think it would be great and exciting to see a lot more like smart people just get into robotics and, and really push that field forward because I think, I think it is hard and I think we’re still in the early innings.

But I think like, if you think about what’s the next trillion dollar company or multi-trillion dollar company, it’s likely going to be in robotics. Awesome. Well, we’ll end on that note.

Thank you so much for joining. And if you haven’t checked out the website, go check it out. Really beautiful videos on the side.

Give you a good handle on what they’re up to. And thank you very much for coming on. And we’ll have you back.

Love to have you back in a year and hear how it goes. Yeah. Thank you, Bill.

I super appreciate it. You really enjoyed the conversation.

LEARN MORE: Discover the essentials of IoT connectivity with our comprehensive guide—an invaluable resource for anyone looking to navigate the complexities of the IoT landscape.