Bill Kerr: What do salad and ice cream have in common? When you order them in a restaurant, these dietary opposites are typically prepared by people and assembling them calls for a bit of dexterity. Until recently, it was impractical to automate such food preparation, but powerful advances in robotics are changing that equation. Welcome to the Managing the Future of Work podcast from Harvard Business School. I'm your host Bill Kerr. Today, I'm speaking with Anthony Tayoun and Dave Johnson, cofounders of Dexai Robotics. Dexai was spun off from the R&D firm Draper in 2018 and recently raised in 2020, a lot of venture capital funding for growth. Dexai's using machine learning to bring robots to commercial kitchens and the fast-casual food counter. We'll discuss today, the logic behind the strategy and consider the wider potential of intelligent robotics. Welcome, Dave and Anthony.

Anthony Tayoun: Thank you, Bill.

Kerr: The coolest part of Dexai, candidly speaking is not the two of you. It's Alfred, and unfortunately, a podcast lacks visual displays. So, tell us a little bit about Alfred and what are some of the basic use cases?

David Johnson: Alfred is a commercial robot arm that sits on top of a counter in a commercial kitchen or fast-casual restaurant. Alfred has a camera and manipulates the same utensils that you see in every kitchen. So, think ladles, spoons, tongs, dishers, spatulas, anything that's used to prepare the same great delicious recipes that you already know and enjoy. But Alfred is able to do it all on its own. It uses artificial intelligence to recognize the different foods and scoops and picks the proper ingredients to make everything from a banana split to your favorite poke bowl.

Kerr: Okay. Well, it sounds like Alfred's versatile for more utensils than I am in the kitchen. So, he's got that in advance. Anthony, if you tell us a bit about why did the multiple end effectors—the multiple tools—why are they important and is that something that's easy or hard for Alfred to handle?

Tayoun: In the food industry, this comes from a regulatory requirement whereby through [the National Restaurant Association] ServSafe [certification program], you need to switch utensils per allergen group for different ingredients. So that's what you see with people doing today by choosing different utensils for the different ingredients. At the same time, you want different ingredients to be handled differently so that you preserve the quality of the food and the taste profile. For Alfred to keep the food preparation process the same as it is today, we want Alfred to be able to switch different utensils on the fly.

Kerr: We of course, in manufacturing and other applications, have seen robotics for many years do one single thing just over and over and over again. You don't have to switch something out. It's the same precise and repetitive task. What's making this possible for Alfred to be able to switch across tasks or to use different utensils one after another?

Johnson: So, the really cool thing about working with food is you never quite know how it's going to show up or what it's going to look like. So, you might have a pile of sliced tomatoes that looks slightly different from what you saw yesterday. And that's what kind of makes this hard in that the robot has to decide what to do on the fly. Our advances in the algorithms really actually enable Alfred to be flexible and adapt to the changing environments. A kitchen is one of the craziest places that you can imagine. It's incredibly busy. Things are flying and moving all over the place. So that flexibility and adaptability is crucial to enable us to actually work with foods and prepare meals.

Kerr: Yeah. This ability or the property that tomatoes will always look a bit different makes this quite challenging. Is it easier to solve that problem in what you are sensing and then the computer sort of adjusting or do you partly solve it through the end effectors? How do you balance between those two approaches?

Johnson: Yeah. We tried the traditional sort of easy approach first, which is let's engineer the environment as much as possible. We can make a specific machine for tomatoes and make another one for guacamole and a third one for cubed chicken. But that turns out to not scale very well, and it also required changing the recipe. One of the things that is crucial to this is we don't want to change the recipes at all. Right? They're already what taste amazing. If I go into my favorite poke place, I want the same poke. It's got to taste exactly the same because that's what I like. So, we had this realization that if we put the challenge into the software instead and use the same utensils, which are already being used to prepare the meal, that means we don't actually change anything about how the food is prepared. But it does make the software quite a bit harder, which is a part of the fun.

Kerr: Yeah. Well, I think I have some possibility of scooping ice cream. I also think that I can possibly make a salad, but a poke bowl seems beyond my paygrade. But Anthony, tell us what has to happen for Alfred to make a poke bowl? This is one of your live sort of product spaces right now that you're rolling out with customers. What have you learned as you've tried to make this dish?

Tayoun: First of all, the robot picks up the order details from the point-of-sale system, so it knows exactly what's being ordered, what are different portions, what are different ingredients that need to be assembled. Then the robot, which has a camera at the wrist, looks into the space and tries to find this ingredient. Once it locates it, it tries to see how that ingredient is laid out. Also, the ingredients are fundamentally changing their behavior in how they're laid out all the time and so the robot needs to adapt to that. Once the robot can see that ingredient closely, it can figure out what's the best way to manipulate it. Then it picks up the correct utensil, manipulates the ingredient, puts it in the bowl and then passes the bowl down the line.

Kerr: Okay. What's the hardest part of this sequence? Where have you guys had to invest the most in terms of the brain power?

Tayoun: First of all, the vision aspect. Second, the path planning, so going through the robot from point A to point B. Then lastly, being able to scoop or pick at the actual ingredients. We employed quite a bit of artificial intelligence and machine learning in each of these steps. I would say the most difficult one by far is being able to manipulate the actual ingredient, particularly because these ingredients change physical properties as you interact with them and that's just something…

Kerr: Meaning, you can squash it, or you can mess it…

Tayoun: Exactly. So, you can squash it. It can fall. Think of just the range of things you can imagine from rice to guacamole to ice cream, how the different consistencies, different profiles of physical behavior and you don't see any other automation product that can do that, right? If you think of the warehouse problem, for instance, you're dealing with pallets and boxes. Things have defined physical structure. In food, this is just fundamentally different.

Kerr: Now, you've set up though an environment that is some helpful in that you've got the point of sale system that's feeding the information in. I'm sure you have to talk to this with potential customers all the time, but someone wants to come in and they want a little less sauce, or they want a little bit more sauce or they want something put over to the side versus on... Can Alfred handle those types of requests?

Johnson: As long as you can define it specifically, no problem. Similarly, Alfred is able to put all the ingredients in very specific places. So if you want it on the side, again, no problem. That's an easy task. What we haven't gotten working yet is you can't talk to Alfred. So, you can't say, "Hey, I want a little more meat," but that's coming in the future.

Kerr: It actually sounds probably a task that you or other people, it'd be pretty easy to handle in the near future of voice recognition and voice response systems. Siri and everyone else is making such advances in that area that that seems like the easier part compared to some of the other things you've been describing.

Johnson: Which is amazing. That shows how far we've come in the last decade. But yeah, we're going to build integrations on top of all those already existing platforms.

Kerr: As the poke bowl gets assembled, as it gets closer to the end, is there a way that Alfred is doing quality checking or assurance to make sure that this matched the original order or are there fault tolerances that it says, "Oops, something's wrong here"?

Johnson: Because it has the camera on the wrist of the robot, it's able to see the ingredients, not only in the hotel pans where it got them from, it's also able to see them in the container where it put them. We record all of that with a video. So not only do you get…

Kerr: You record every poke bowl?

Johnson: Every poke bowl. So that's the amazing... It's one of the data side effects is that Alfred has a copy of absolutely everything it did. We use that not only for quality control, but also for learning. So, every Alfred gets the benefit of every action that every other robot took, which is this amazing ability to learn across our fleet of robots.

Kerr: Yeah. So, I just want to just emphasize that because I think it's very important, which is historically, we put the technology into the robot and we put it on the manufacturing floor and it just did what it did. It didn't learn, or anything that maybe it learned, it was localized. But you're using all of the Alfreds throughout the world to constantly update your learning bank, this data repository.

Johnson: Exactly. So, every time Alfred scoops guacamole, every other one learns how to do it slightly better.

Tayoun: And similarly, when we have new tasks that are being rolled out, when one Alfred learns a new skill, every other Alfred learns that too. Whether that's flipping a burger, operating an oven, whatever that task is. And the benefits for-

Kerr: I'm sorry, just to interject, and that could possibly say that if I have an Alfred installed at my restaurant and I do, in fact, want to change the recipe, and you, Dave, began by saying we don't want the customer to need to change the recipe. If I said I do, then I possibly can add in new skills or new tasks to my line based upon what Alfred's learned in other places.

Tayoun: So, that can happen in a lot of different ways, and that's one of them. Another one is we can train Alfred in our own kitchen at and then roll it out to all their customers if they need that new skill. And yes, as an operator you have the freedom to choose which skills you want. You can edit the recipes live, you can tweak the portions, you have absolutely full control of your whole process.

Kerr: Why food? That's not the one that was so obvious, I think. And I think even three years ago when we began the Managing the Future of Work project at Harvard Business School, we would not have thought of robots in the food services industry any time in the coming probably five years or even decades. So, what changed, or why pick food?

Johnson: Well, to me food seems completely natural. So, if you ask any of my friends, I'm a bit of a voracious eater and a hobby chef. But on the other hand, the food industry is under a really tremendous labor shortage right now. There is this massive rise of delivery services, and food-on-demand is this incredible new domain, which is really driving a lot of consumer change. That is also combined with the fact that you have far fewer teenagers in the workforce, and a lot of competition for the labor that has traditionally worked in restaurants. So those factors, when you put them all together, mean that there's a dramatic labor shortage. And restaurants are really—75 percent of them or so are understaffed. The traditional solution to that kind of labor shortage is automation. And that's really the problem that we set up to solve.

Tayoun: And from a technology standpoint, a lot of these things were just not possible before. So, from Managing the Future of Work, we saw that a lot of times food service automation is among the top sectors to be automated. And that's just because of the way the nature of the tasks are, in terms of repetitiveness, in terms of knowing what needs to happen. And only recently were we able on the technology side to automate, if I may say, semi-repetitive tasks that are happening in non-structured environments. And that's because of a lot of advancements recently in the computer vision side and in terms of computational availability on premise. And at the same time, robots became much easier or much finer to control. And so, all of these things coming together made it possible just now to be able to automate this big set of tasks.

Kerr: Yeah. So, you're highlighting here that, I think you said semi-routine—that every salad bowl is going to be modestly different, and Alfred's going to handle both the different ingredients that come in, as well as also whether or not someone wants croutons or other on their package. But at the same time, it's not this free-wheeling environment where everything—you're not asking Alfred to pick up 43 different ingredients and make for the first time some random dish that somebody is asking for, on there.

Johnson: Exactly. And we're also not even starting from scratch. We don't start from dirty vegetables, for instance. We're starting now from prepared foods. And that makes a big difference. Food prep is something that's on our roadmap to do in the future, but it's these initial steps which are now possible. And one of the things that we discovered is that when you automate recipes, you add an additional layer of specificity than what the restaurant originally had, because the robot is going to do things in a very ordered and structured way, just because that's the nature of the automation, even flexible automation. So, we're taking a task which had even more variability than you might expect when we started, and making it be extremely precise, which adds additional quality control and uniformity to the final product. But food is not something that's had that kind of strictness to it before.

Kerr: So just to make sure I get that, if Bill was hired into a McDonald's or into a Chipotle or a Qdoba or something like that, I'm told this is how you're going to make the burrito bowl. But I figure some things out, or I see what my neighbor's doing and I follow that pattern. And you're making it much more, “this is exactly the amount of beans that you put into the burrito each time.”

Johnson: Exactly. And it may be that you had a set of instructions previously, but you never followed them perfectly. And that's just the nature of how things are done. And Alfred follows the instructions perfectly every time.

Kerr: Because you think about a restaurant that's already up and running. So, I have people that are doing the work of the salad bowl development line, or they're sitting in the back doing the food prep for the ingredients. How can they think about putting Alfred into their work? Is it that Alfred should do all of the food prep, or all of the food assembly, or can you insert it in part of the role? What does this look like in an existing restaurant?

Tayoun: We've built Alfred in a very modular way, and the intention is that Alfred today can do all of the food assembly tasks. And depending on the throughput you need at a restaurant, you can place several Alfreds next to each other and they'll work together. You can place Alfreds alongside human counterparts and they'll work together too. Alfred is this very collaborative tool that restaurants can use to their advantage. So, we try to think of Alfred as more of an appliance such as a microwave or a dishwasher, and not really any fancy automation tool. And, so for a restaurant you can choose to use Alfred as much as you want for your needs.

Kerr: Okay. So, if you're having trouble bringing the teenagers in to work, many of them today would prefer to go to work for Uber rather than to work in one of the traditional fast food restaurants or something similar. This can be something that you can deploy in a flexible way alongside that workforce need.

Tayoun: Exactly. Exactly.

Johnson: Yeah. Alfred is really designed to fit into existing workflows and take a specific portion of a task or a full task and perform that over and over again, repeatedly and reliably. So this means that if you have a restaurant that has multiple items on its menu, it could be burritos, or poke, or fajitas, or anything like that. Alfred will then be able to do any of those recipes, because it only matters that it has the utensils and it's able to reach the ingredients.

Kerr: So, that's in a setting where we're already up and running and we have an existing process and we're trying to maybe fit Alfred in, at least in the near term, into that process and take on some tasks. If you were to instead be talking to somebody that's building up a restaurant from scratch, like they're in just the original design of the flow of the restaurant, how the food's going to be prepared and so forth. Do you come at it in a different way, in terms of what Alfred could do? And would it matter if it's fully automated versus intentional for partial automation?

Tayoun: We haven't done that yet, but that's a very, very exciting thing to do. Specifically, you can think of new ways to utilize your space with Alfred. So, one of the examples is, Alfred doesn't care if it's scooping from bins that are stacked vertically or horizontally. Alfred doesn't need to walk around. So, it could be surrounded by food from all sides. And you can already imagine a better utilization of your space as a restaurant, a better flow for consumers, where they can see all of the ingredients laid out in a much more efficient way simply because there are very many requirements that the robot does not need. And, so as a restaurant operator you'll get much more flexibility in the way you use your space and the way you have the workflow for the different customers and the way the experience happens more generally.

Johnson: Yeah, we like to think of Alfred as an opportunity for restaurant operators to focus more on the hospitality aspect of their business. So, Alfred takes a lot of the repetitive assembly and food prep tasks away and then enables the restaurant staff to focus on interactions with customers, improving the customer experience, and everything that really makes the in-person dining experience so valuable.

Kerr: Yeah. One can imagine a future, I know many restaurants and chains are preparing for where a lot of their orders are going to come in on the app beforehand, or some other form of a kiosk. And then there's the food preparation. There's going to have maybe fewer people, but they're going to be in a much higher touch customer facing role on the outside, as well as you described earlier, food preparation. Someone still needs to fill the poke bowl in front of Alfred that he's going to be a scooping from. But you also are in dark kitchens. And I think that is, in my understanding of the food service space, the fastest growing part of the industry. And tell us a little about what, first off, for many people, what is a dark kitchen? And then why it's becoming more popular, and then what are you guys doing in that space?

Johnson: A dark kitchen or a “ghost kitchen” is a non-consumer facing, retail kitchen which prepares food that's ordered exclusively on delivery services. And these locations will prepare multiple different brands or different cuisines and they are picked up by the drivers and then delivered to the consumer. That has huge advantages for scale and cost reduction because you can combine so many of these different food types and brands into a single location.

Kerr: And Dave, give us a sense of how big is a dark kitchen, relative to the typical McDonald's that we'd walk into and where would they be located?

Johnson: Oh, they can be really of any size and in any location. Which is one of the beautiful things. Because they're not consumer facing, you can put them in real estate that's not high foot traffic, but actually has great access for the delivery drivers. They can go in urban centers to be close to a lot of consumers. Or they can go on outskirts and then be delivered to the consumer via the drivers. That flexibility means that the real estate costs can be substantially lower and the cost savings can be passed back onto the consumer.

Kerr: And how are these multiple cuisines being aggregated into one spot? Is there an owner of the dark kitchen that has aggregated the separate contracts with the various cuisines? Or is there ... How does that part come about? Or is somebody renting out space to the cuisines?

Johnson: So, there's a bunch of different financial models—all of those that you described—so that you can either rent space if you're a brand owner and then use that as a location to produce your food. Or you can have an operator which will contract with the brands and they will then prepare the food and take care of everything for you. We haven't seen which way really is going to win out in the end, which is what sort of makes this whole time really very exciting.

Tayoun: In a traditional restaurant, you have your bottleneck is really the number of times we can turn the tables that you have in the seating area. Whereas in a ghost kitchen, by removing that seating area and by opening up to all the delivery services you have, your bottleneck becomes the throughput of your kitchen. And, so the more you can increase the throughput of that kitchen, the more you can generate return, which really pushes up your return on your assets, your return on the real estate that you have, and your return on any of your costs, really. And so that makes the financials of the business just infinitely more attractive relative to a regular restaurant, provided you can keep the demand higher than the throughput you can meet.

Kerr: This again is a phenomenal force that's entering into the food service space. Now, let's go back to Alfred. What does Alfred do in these environments? Is it an easier environment for Alfred compared to working in a customer-facing restaurant? What's the best part of the value propositions here?

Tayoun: So, if you think of Alfred also as an asset, then by utilizing this asset more, the return for everyone who's using that asset increases significantly. And since ghost kitchens have this characteristic that they're opening up the demand and they're improving their utilization, then this naturally flows to Alfred as well. And, so ghost kitchens became this natural home for Alfred, where Alfred can thrive by increasing the utilization and increasing the positive effect on both the consumers and the end users.

Kerr: And just to remember back to your little data piggy bank—also, the more Alfred's doing this stuff over and over and over again and one of these very high frequency environments, the more the algorithms are learning.

Tayoun: Exactly.

Kerr: So, this brings us to a point we haven't yet discussed, which is how do you sell Alfred? Like do you sell Alfred as like this bot that's got some price tag on it? Do you sell it like on a daily basis? What's the way that that someone pays you for Alfred services?

Tayoun: The way we commercialize Alfred is through robots-as-a-service as a model. And that means that our customers don't have to pay any capital costs for hiring Alfred. Instead, Alfred this charge solely on a revenue share basis. So, for every poke bowl, every salad, every item that Alfred does, we earn a portion of that revenue. And to give you an idea of scale, the portion that Alfred charges is significantly less than what the delivery services are charging the restaurants today. And they charge on a similar revenue share model as well.

Kerr: Okay. So, this is important for the restaurant owners kind of approach, because you've taken what is a semi-fixed costs in I have an individual there for shift work or something like that. And now you've made it a pure variable cost in their calculations.

Tayoun: Exactly.

Kerr: And we've obviously seen software-as-a-service become something quite important in our economy. And robots-as-a-service is now a new form that would be provided on a continual basis to people.

Tayoun: Absolutely. And as robots continue to be commoditized, we're going to see more and more robots-as-a-service model, it's coming up.

Kerr: So, as we think about the ... some of the frequent questions I'm sure you guys have handled—but let me raise them. One would be like some liability or health concerns, that you've got now this robot that's kind of making the food and maybe that's good or bad. Like talk to us a little about relative to Bill Kerr making your poke bowl, is Alfred a better choice? And let's take the skill out of it—of me not being able to chop fish—but go towards the health and safety concerns.

Johnson: Alfred is certified to be safe for working around humans because it's a cobot. So, Alfred first off has torque limits so it's not allowed to exceed certain torques. And through its cameras it's able to recognize people in the workspace and it always stops as soon as a person approaches, which means it's a very safe appliance that isn't able to actually be moving while a worker is within the danger zone or is close to be able to getting hit by Alfred. As far as Alfred being food safe, there's really three main ways which it's a much better solution than what's already being done today. So, first, Alfred is certified by the NSF [NSF International, originally the National Sanitation Foundation] as being a commercial food safe equipment and, so that it meets a cleanliness and clean-ability standards. Second, Alfred is able to measure and has a lot of data about the actual ingredients that are going into the food. So that means that the food you know has met the time and temperature requirements. So that it's not been out too long and it's always been maintained at a safe temperature. And finally, Alfred takes out a lot of the messiness associated with people actually preparing the food. So, Alfred is always clean. You never have to worry that Alfred didn't wash its hands and Alfred doesn't have to deal with any of the problems associated with dirty uniforms or hair nets or anything else like that.

Kerr: Unfortunately, as we are recording this podcast, we are in the midst of the coronavirus and it's spread across the US, it's March of 2020. So, in that environment where you're really concerned about viral diseases and infections, is this going to be something that's more important for us to have in future food service prep?

Tayoun: So, that's the other benefit here is that many of the foodborne illnesses that are being spread are spread mainly because of humans spreading them around. And the same is true of the viruses such as the coronavirus here. And with Alfred, you never have to worry about that anymore. Because Alfred is this fully sanitary solution that is handling the food without spreading any contaminant at all. So, as long as you're sure that you're a value chain does not have any contaminants, then Alfred will never add any external factor, any contaminants of that nature to the food.

Kerr: Yeah. The other thing you've also mentioned along this podcast is that you are recording all of this food preparation. So, every taco bowl that goes through in front of Alfred, you have a record of that. Does that factor into these legal questions? Is it a safety kind of check that you can run with? What could you do with that information?

Tayoun: Oh, certainly from a consumer standpoint, if you ever have any question marks about how your food was prepared, or if you have any accusations, or if you have any follow-ups that you want to make, we can send you a video of your actual meal being prepared so you can see precisely every bit of ingredient that went into your meal. You can get a video of that. You can get a list of all the ingredients and the precise calories that are going into every meal that you're eating, which gives you this great control and information about what you're eating that previously was not possible.

Kerr: Alfred has a lot of skills already, and when you think about your ongoing development path in R&D, where do you see Alfred going next, like over the next year, next two years in terms of the types of food that he's going to be ready to handle?

Johnson: So, Alfred really is going to go everywhere in the kitchen. We see Alfred as this universal appliance that's able to first be able to handle the grill. So, think grilled chicken, steaks, burgers, anything that is cooked by flipping it on a grill. Then working with ranges, so we can sauté or pan fry, work with woks. Then moving into other appliances, so handling ovens. A great application is cooking perfect roasted sweet potatoes for instance. Alfred can put the pan into the oven and remove it after the appropriate amount of time. Then we're going to do food preparation. This is basically taking material and putting it into a variety of food processors, eventually slicing and chopping. There's really an amazing number of actions that this flexible and articulate robot arm can perform.

Kerr: Would you ever have a vision for Alfred becoming mobile? Obviously, the arm swings around, but like a track system where Alfred could go up and down and do something at the oven, come back to the griddle, flip some stuff, go back to the oven.

Johnson: Yeah, mobility is this really cool frontier for robotics. I think we're going to explore a lot of those kinds of solutions. Because we're a software company really built on top of these commercial robot arms, as these mobile platforms become more ubiquitous and more available Alfred will look slightly different, but its skills will be transferred to these new platforms. The beauty of our approach is that it doesn't rely on a specific physical implementation, and it can be adapted to basically anything as it comes out on the market.

Kerr: As you think more broadly about the service industry space—and this goes beyond Dexai directly—if we look ahead to 2030 do you think we're going to see robots everywhere in the services sector? Is this kind of like at an inflection point or how do you forecast that future out?

Johnson: Automation is this amazing thing that you don't quite appreciate it until you really see it. But as these robots become more mobile, a little smarter about their environments, they are just going to go everywhere. Manufacturing saw this over the past several decades, where robots have effectively taken over nearly all of the repetitive and automatable tasks in manufacturing. We're going to see the same thing in these more semi-structured environments, and then eventually in unstructured environments.

Tayoun: And we already see that with the availability and the advancement that's happening on the hardware side. As we mentioned in the beginning of this podcast, these robot arms are becoming very commoditized, which means it's opening up the door for a lot of software applications to be built on top of them. The same thing happened with industrial robots 50 years ago. The same thing happened with smartphones when they first came out. You have this hardware that becomes a commodity and then you open up really the door to a lot of software innovations. What's really unique here is at the same time we're seeing these huge advancements in machine learning and artificial intelligence that are also setting up the stage for a lot of new applications within robotics.

Kerr: I think that's one of the reasons why we have found Dexai so head-spinning is that this would not have been the use case we were imagining three years ago. And yet you're clearly not only identifying it and working on it, you're already deployed in a number of installations around this. The other driver that you mentioned at the beginning was labor costs and the challenges of finding workers, and as we have a population that continues to age and the workforce shrinks, being able to... That labor scarcity challenge is not going to go away any time soon. You've probably heard of robot taxes or proposals for robot taxes. Does Alfred have any views about being taxed?

Johnson: I think your dishwasher also has views, or doesn't really have all that many views, about being taxed. We think of this really like an appliance. Maybe a dishwasher tax makes sense. But I think it's sort of in the same vein that, you have this tool that's able to perform a repetitive task. It's really not doing the innovative or the creative part of the process whatsoever. So, it's more like closer to Photoshop for chefs really than anything else, in that it's able to take the inspiration and the creativity of somebody and turn that into a delicious meal, but really in the same way that a dishwasher or a microwave [works].

Kerr: All right, two great entrepreneurs. Anthony and Dave, thanks so much for joining us today, talking about Dexai, really how artificial intelligence and advanced robotics are coming together to do some astounding things in the food service space, and what we're probably going to be seeing a lot more of over the decade ahead. Thanks guys.

Tayoun: Thank you.

Johnson: Thank you.

Kerr: Thank you for listening to this episode of the Managing the Future of Work podcast. To find out more about our project on the future of work, visit our website at hbs.edu/managing-the-future-of-work and sign up for our newsletter.