Building an Autonomous Delivery Experience with DoorDash Co-Founders Andy Fang and Stanley Tang

Hi listeners, welcome back to No Priors. Today I'm here with Andy Fang and Stanley Tang, co-founders at DoorDash. We talk about how you can ask DoorDash in natural language for food and groceries, what that means for the future of agentic commerce, their delivery robot, Dot, how DoorDash has been a robotics company for the last 8 years, the data advantages of their network, and what all this means for 9 million Dashers and 3 billion deliveries a year. Welcome. Andy, Stanley, thank you so much for being here. Really excited to talk to you about all the crazy stuff DoorDash is doing. I thought we could start with what's going on with Agentic Commerce at DoorDash. I feel like you have one of the largest rollouts of actually using AI to change what people consume. Yeah. So what was the backstory here?

I mean, it started a couple years ago. Honestly, in terms of like our attempts to try to make a play here, it actually— originally we were bullish on voice as the modality. Um, that—

and then ended up not being the thing.

That ended up not being the thing, but maybe it will in the future, but it just— that didn't really land. But the thing that was very interesting for us was just this natural conversational experience. And I think, you know, what we've seen is just like people being able to, like, just, like, naturally just translate what's in their head into this interface versus trying to, like, do some research online and then try to do some, like, keyword optimization stuff. Like, people just found it easier to search for things, either more nuanced kind of restaurant discovery searches or different tasks on the grocery side. Um, and yeah, we've just seen a lot of interesting traction that's upheld as we've expanded the rollout.

What are you seeing in terms of behavior change from the user side? Like, do I eat or buy differently?

Yeah. So I I would say on the restaurant side, we are seeing people, 50% of trajectories of people using Ask DoorDash for restaurants, uh, they're or 50% of those trajectories are people ordering from places they've never ordered from before, which is huge because that's one of the hardest metrics historically for DoorDash for us to move. And so that's been big. And then another one is on the grocery side, we're seeing a lot higher basket sizes, like I would say like 40% larger basket sizes on grocery. And so people are like, you know, they'll take a picture of what's in their fridge and they'll say, hey, help me stock up my fridge, or they'll do meal planning with like, maybe they have some dietary constraints, or they're like, hey, I want to like cook a pasta dinner this weekend with my family, or even just like, hey, help me reorder, uh, like my, you know, my usuals. And like, that's a lot easier than tapping through the traditional experience.

That's wild. I've never thought of DoorDash as difficult to use, but like, that suggests there's like actually latent demand that wasn't being served because you— it wasn't easy enough to like eat at new places.

Correct. Yeah.

And I think a lot of people on the restaurant side, it's like people build habits, but I think people also want some diversity in terms of like what they're eating, you know. Um, and so we felt like this experience ended up being a natural way to allow people to express that.

Oh, think about the social currency of like, my friend Andy found a new, like, really good restaurant for me. Right. Yeah. Andy's awesome.

Right.

So I feel like that's even a different way people look at DoorDash.

And yeah, another thing that was an investment we made was actually like incorporating bringing like world knowledge into the experience.

So what does that mean here?

Things that are going on with restaurants outside of DoorDash. So like, you know, we'll see, hey, what's trending on the internet or what's stuff that's not in the models, but stuff that people would find because like their knowledge cutoff is too early. But maybe it's like, hey, what's trending online or what are people talking about in various forums or whatever? And kind of goes to your point of like, hey, like kind of want to eat what's cool. And so like that was something we tried to incorporate into the experience to make people trust it.

More. How do you think, uh, people will buy or think about restaurants differently in like 5 years from now?

I don't know about 5 years from now.

I realize it's really hard in the age of AI. Like, next step.

So for Ask DoorDash, I would say to start with, maybe that's like the next couple months or so. I think it's making it easier for people to discover the experience and like figure out what to do, because I think it can be intimidating if you just see like, hey, like there's like suggested queries that you can type, but like Some people don't know what to start with. So figuring out how to experiment and tinker with the user experience to kind of get people or encourage people to find use cases for it. I think if I think further out, then it's a little more speculative. But, you know, Stanley and I talk about this all the time. It's like if someone were to create DoorDash today, like, I don't know, like college kids in a garage trying to start DoorDash, I think it would look very different, probably more agentic first. You know, one stat that I always like to, uh, think about nowadays is just like there's more agent traffic on the web than human traffic, you know? And so it's like, how do we have a DoorDash type experience that plays into that trend? Um, and so, you know, I think there's some interesting speculations there, but hard to say.

What could my agent know about what I want to eat or what I want to, um, buy from a grocery perspective? Like, help me understand like how you think about richer context or how to be smarter there.

Sure.

I mean, One cool example is someone's like, hey, for our office, it's like, I can just like have the, like one of the cameras on the, uh, pantry shelf.

Mm-hmm.

It's like, hey, when the shelf starts to get empty, like I can fire off like a query to DoorDash to like stock up my shelf.

Yes. As a human being has care. Yes.

Yeah. Yeah.

And so that was kind of like, I mean, something we talk about more later, but like kind of our, like early experimentation with our CLI is like, that's kind of an example of like making it less friction for an agent to kind of, like, participate in that experience.

Okay, well, while we're here talking about user needs, I've got to be, like, a top percentile DoorDash consumer. Nice.

I know.

I'm, you know, a lot of customers at this point. But I host family dinner for, like, extended family every Sunday night. And, you know, we eat DoorDash because I'm not going to cook for all these people every week. Or I can't all the time. Um, and, uh, like I do the same thing every time, which is poll everyone. Okay. Who's coming?

Oh yeah.

And then these people have these allergies and whatever else. And like, does anybody feel like anything special? Yeah. And then, you know, I order, right? And I'm like, I feel like, I feel like that's all within the realm of possibility.

That is definitely a use case.

You just put it on autopilot for me. I show up, I hang out with my family. Everything's good.

That is a use case that is, I mean, I think, uh, not exactly the same, but like a similar use case is like the office lunch ordering kind of thing. It's like if you're the office manager, it's like, I don't want to like— and then you got to like, hey, make sure you ordered lunch at this time, otherwise it's not going to show up. And it's like, again, everyone has their own like allergies or dietary preferences and stuff.

So Stanley, you guys are doing a whole bunch of things on the autonomy and robotics side as well. Like you're clearly— your view of DoorDash as founders is broader and more ambitious than, I don't know, maybe just like the surface-level view of it's a food delivery network, or what, whatever the first, you know, one-liner for the company was. Um, how long ago did the robotics efforts start?

Yeah, we've actually been looking into robotics and autonomy probably much longer than people thought, like since 2018 actually, uh, back when it wasn't obvious autonomy and robotics was going to be a thing. But we felt like this was going to be a technology that was going to be transformative to our space and potentially disruptive. And I think, I think that's the nice thing about being a founder-led company is like we are, we get to think about kind of much more future speculative things that are on the horizon and, and, and constantly think about like, how do we make sure we don't get disrupted by the next one. I think like, like Andy said, like the next DoorDash that comes along is not going to be someone that builds the exact same version of DoorDash, but maybe with a better UI. It's going to be—

that would be dumb.

Yeah, it's going to be like something like, okay, how do we incorporate AI into commerce? How to incorporate autonomy, robotics, drone deliveries, etc. And I think, I mean, fast forward like 7, 8 years later, I think you're seeing everything starting to play out in AI. In robotics and autonomy, seeing Waymo happening. And I think, you know, we're glad that we made that investment early on in 2018.

DoorDash is an amazing business. In 2018, it was like less amazing than it is today. I feel like that's a fair statement, right? How do you think about like the timing and sequencing of these very long-term bets and like just from a capital allocation perspective, like when you can invest in these things?

Yeah, I think it's probably the same of how we invest in a lot of things at DoorDash is everything started out as experiments. I mean, in a way, that's the— that was the founding story behind DoorDash. DoorDash was a Stanford College, like, dorm room experiment. It started out as a website called PaloAltoDelivery.com with 8 PDF menus and a Google Voice phone number. And it was only once we figured out, okay, there's something here, let's turn this into a company. And that's basically— we've kind of taken that philosophy throughout the past 13 years and we've kind of applied it to autonomy as well, AI as well. I mean, when we first started in 2018, the intention wasn't, hey, let's go spin up this giant robotics program, let's hire roboticists, go build hardware. It was really— we put together— it was me and half an engineer's time. It was a skunkworks project, the experimentation to go, let's go explore, like, what's out there. Like, we don't even know what autonomy looks like, how robotics is going to impact our space. But let's go explore. Let's go form partnerships. Let's go learn. Let's go experiment. And, and, and I think in the beginning, The intention wasn't to build our own robot. Actually, we didn't think we needed to build any of this technology ourselves. We thought, okay, we can just partner up with a bunch of folks. Like, you know, back then we weren't, you know, we didn't know anything about robotics. There's all these startups out there that have built robots and autonomy. Like, why don't we just work with them? We can essentially just be the platform. We'll build the APIs, we'll handle the distribution, etc. And we did that for about actually several years actually. We worked with everyone in the space, everyone from the sidewalk robot players all the way up to the robotaxi players. I'll say there's 3 things we learned through that experience. I think one is it kind of validated or confirmed our belief that there's something here. Autonomy is a question of when it was going to happen, not if. And again, fast forward today, you're seeing— you see the Waymos driving, right? It's happening. So we should keep investing. The second is I think it allowed us to learn what it takes to actually enable autonomy. Because it turns out there's a lot of things you have to build around autonomy, the infrastructure, the ecosystem. How does autonomy integrate with DoorDash? What deliveries do you take on? The operational aspect. It turned out a lot of things you have to build around autonomy in order to make autonomy possible. It's not just you plop a robot in, or even AI, just plop an LLM in, and then things just magically happen. There's a lot of things around it, and you kind of have to build a platform ecosystem. One of the things that we ended up building is this thing called the Autonomous Delivery Platform. Essentially, it's like, what are all the products and technology, the APIs, the dispatch you need to build now that in a, in a post-autonomy world where autonomy and robotics and drones are everywhere, what are all the things you have to build? How do you integrate with merchants? What does the consumer experience look like? And I think the last thing, which I think is probably the most important thing, we learned, which eventually led us to realize we had to build this technology ourselves, is really this idea of building towards a use case. Yes, there's a lot of autonomy startups out there, but it always felt like these, these companies weren't really focused on a use case. It always felt like they kind of build the technology first and then retroactively try to go find a problem to fit into. Like these, these things are all built in a vacuum, which is kind of weird because, because it's like because in software software world, like when we went through YC, like we're always taught to, oh, you got to start with the customer, build something people want. That's kind of like drilled into you and then you can iterate. But then when it comes to like hardware and hard tech and AI and robotics, it's people just kind of do the opposite where they try to build the tech first and not really think about the use case they're building towards. And whenever that happens, you just end up with something that just wasn't quite the right fit. Like, like there's— and we went through this, this, this process where a lot of these companies out there, but I always felt like it wasn't exactly what DoorDash needed. Um, like, like for example, you, you, simple example is that you have these in, in, in Tonny World, there's basically two buckets of category of, of companies out there. You have these sidewalk robot companies, which are kind of these 2, 3 mile per hour kind of water cooler on wheels, super effective, simple technology. Uh, but we quickly realized the speed was like, and distance was a huge limitation. 'Cause if you— 'cause the average delivery at DoorDash is about 3 to 5 miles, uh, and, and, and the typical delivery time's about 15 minutes if you exclude the time it takes to make the food. So if you put a 2-mile-per-hour sidewalk robot, it's just never gonna work. And then on the, on the end of the spectrum, you have kind of the robotaxi players, which really are designed for carrying people around. It's a 4,000-pound vehicle. This goes super fast. You're transporting people. And, and, but it turns out the problem around carrying people and carrying goods is actually a little bit different. Like, you don't need— if you only have— if you're only carrying a couple burritos around, do you really need a 4,000-pound car with chairs and AC? Uh, the pickup-drop-off problem is also very different in robotaxis. Um, you know, you can walk to a Waymo. I mean, how often have you taken a Waymo where it drops you off half a block or a block away from where you need to be? Which is totally fine because you can, you can walk, but Packages can't do that. Like, how do you, how do you solve that? What I call the first and last 100 feet problem. How does the food get picked up at the merchant? What does that integration look like? And then on the customer, like, how do you drop off the food? How do you find the driveway? Like, people expect their food to be dropped off or the vehicle to be pulled up straight to the front of their driveway or their, or their porch. So, so, so we kind of looked around and at and asked ourselves, okay, like if you were to start first principle, then again, this has always been our philosophy at DoorDash. Like, like if you were to start from the business, the customer use case, work your way backwards, our first principles, and you can build exactly what we need to solve our use case. What would that look like? And we looked around. Turns out no one's really building that. It's not a sidewalk robot. It's not a robotaxi. We felt like it was probably something in between that the right metaphor for us. Again, it's like if you're trying to solve that 3 to 5 mile delivery in dense suburbs, which is where most of the deliveries happen, the right metaphor is probably an autonomous motorcycle or scooter or bike profile vehicle. And, you know, it doesn't even be 4,000 pounds. It's probably, you know, 300 pounds, but also has to be a lot faster than Sidewalk Robot, has to go 20, 25 miles per hour.. And when we looked around and saw no one's building that, we decided, well, if no one's going to do that, instead of waiting around and let's wait for this to happen, we're going to control our own destiny here. Let's invest in this and see what we can build. And it took many iterations, you know, like we start looking at like when testing this with real DoorDash deliveries, looking at our 10 billion deliveries we've done, extracting the insights we have on the operational learnings we have. And that's eventually what led us to launch and ship, which is kind of our in-house autonomous delivery robot. So it's been a— it's been quite a journey. And but again, this is something we look to bring to every aspect of the business, whether it's autonomy, robotics, AI, like it's always starts out as experiments. It always starts out as what is the customer problem you're solving for? What's the use case you're solving for? Work your way backwards and then iterate and validate kind of your hypothesis and slowly build the product over time.

That sounds extremely rational. I have a hypothesis and it's very cool. I want to ask you where we are in the life cycle of everybody getting these automated deliveries. Um, I have a hypothesis and I'm curious if it resonates with either of you about like why, uh, a lot of people in this era are building technology first versus customer back. I think people think everything is going to work like ChatGPT. Yeah. Right? I just, and like, by the way, like, there was, of course, work done on, uh, instruction fine-tuning to get it to, like, be shaped in a product that was still a user experience. But I, I think the, the mental model that people have of, like, it's a general technology and it's just kind of, like, free to turn into different applications, uh, is what they're applying to lots of different things now. And especially in autonomy, my sense is people are like, okay, we'll make the model. And then like the other stuff will be, uh, if not easy, at least secondary. This is not my view at all.

I, yeah, I agree with you there. I mean, that's basically your methodology into building the form factor.

I think maybe that approach works in like software land, but like for at least for a business like ours, like DoorDash is a physical world business. It's like, you know, you bring technology into physical world and the physical world is always A lot messier. It's a lot more complicated, a lot more nuanced. Uh, I think one of the things I think people don't realize is just how complicated DoorDash is. I mean, we do over 3 billion deliveries a year. There are no two deliveries that look the same. All 3 billion deliveries look, look different. Uh, and they all come in all sorts of shapes and sizes and different geographies, like a delivery in in downtown San Francisco is completely different than, uh, a delivery done in Dallas or, or even in Europe or in Helsinki where it's snowing. Or if you're doing a pizza, it's very different than ice cream. Like, your dinner is very different than your grocery order, which is very different. Now that we're expanding to retail and, and, and pharmacy and parcels as well, it's like the diversity of deliveries that happen at DoorDash is so complex that I think people sometimes don't realize just how nuanced the problem, the problem, the problem is. And that's kind of how what we have to solve for at DoorDash. And I think that's part of the— been the learning process, especially when it comes to like building autonomy or even AI, is how do you manage through all that complexity? And again, it always comes down to like, like, do you understand the use case? And I think we just have such a huge advantage over everyone else because we have something that everyone else doesn't have. It's called DoorDash. We have 10 billion deliveries of data to extract from. We have all these consumers, you know, over 40 million consumers ordering every single month. Like, we understand the complexities of how to handle when things go wrong, how to integrate across all different types of merchants. Like, the way you work with a McDonald's or Starbucks is very different than working with a mom-and-pop sandwich shop. Like a, a drive-through restaurant is, again, it's very different than a restaurant at a strip mall or downtown Main Street. And how do you handle kind of those different use cases, right? Different interaction, different pickup points. Um, I don't know if there's anything you wanna add on the AI side.

I mean, for me, like the kind of that analogy you brought up, I think, I think about it in terms of the autonomy thing, but I also think about it in terms of like the human, like how the humanoid robotics space is starting to play out. Potentially, where— I mean, we, we also launched a product called Tasks a couple months ago where we're, we're having people in the Dash Fleet help basically collect, uh, data points to help train some of these world models. And I think we're so early there, and I think there's so many different form factors that you can use, and there's like different opinions on like what type of model is going to work versus not. Um, but I think unlike something like ChatGPT, I think there's a lot of expense needed to invest in just like the V1 of this. I guess ChatGPT costs a lot of money too, but I think there's a lot of pressure though to figure out how do I actually provide value? Like I have to be better than what people can do today. Um, and you know, whether it's .NET and like delivering something end to end, or I mean, you probably invest in like a bunch of different players in the space, but like there's real pressure to like be better than the alternative, uh, from either a quality and/or a cost perspective. So yeah.

Yes. Otherwise, what are we doing?

Yeah, exactly.

Um, so for those of us who aren't in Phoenix, like what is DoorDash Dot and like tell us about the design of it.

Yeah. So DoorDash Dot, it's an autonomous delivery robot. It's built entirely in-house, uh, at DoorDash. It's, uh, weighs 300 pounds. Travels up to 20 miles per hour. It's 1/10 the size of a car. It's the only delivery robot out there that's designed to travel not just on sidewalks, but also go on bike lanes, on, and on the road as well. It's, it's live in Phoenix. We've been live doing deliveries for almost 2 years now. It's, you know, we do— it's fully autonomous L4. So if you come out to Phoenix, to Tempe, it really feels like, Waymo San Francisco.

I'm going to state something and see if this is like correct or you agree. Uh, even beyond understanding the wealth of use cases, like, you need to know what the distribution of environments you're going to be playing in is in robotics. This is a huge problem for everybody where like, it's not— I think most people, uh, familiar with the area understand that it's not that hard to get a cherry-picked demo of like one cool success on a task.

Right.

The problem is getting it to work on any object or in any environment. Yeah. Um, and so there's this like, you know, huge question in the industry of like, okay, how are we going to go get data that feels like realistic data? And like the best realistic data is the real world data actually. And so I think that's like a really interesting premise of like why you might have the right to go do this besides you want to do it. For the quality of your business.

Yeah, no, exactly. And I think that's, again, that's also where DoorDash gets to shine with our advantage is we don't necessarily have to solve for 100% of our use cases. I mean, that's, that's also part of our— again, that was part of the learning with our kind of the, kind of the first early years when, when we did the partnerships, right? We built our autonomous delivery platform was understanding what kind of deliveries fits into what modality. And, and I think the vision was always, was always let's not design something to solve everything, but instead kind of let's, let's go with it with a how do you come up with a multimodal strategy where perhaps, you know, you have DoorDash Dot do kind of the 3 to 5 mile suburban deliveries from a strip mall. So, so right now we're live in Phoenix. That's kind of our starting point with Dot. That's kind of the perfect market for these dense suburbs. Yes, things are still far, far apart enough. Maybe if it's, if it's a rural area where there's poor road infrastructure, maybe you send it and it's a lightweight order. Maybe you send a drone delivery for that. If it's a complicated multi-step grocery order, we have to climb, go up and down stairs and pick and pack orders. You're still going to have a Dasher for that. And I think that's the nice thing about DoorDash is you can kind of— you don't have to— it's not an all-or-nothing approach. You can kind of phase in these modalities over time and pick and choose what the right— again, it's about the use case. What are the right use cases to solve for? What are the right modalities to fit into for each of the use cases? Like, are there certain deliveries you can carve out that makes a lot of sense for robotics versus humans?

Yeah. I also think that's really cool that you have control over the routing and the distribution where you're like, I can, I can accomplish this task.

Exactly. And then, and then from the consumer side and the merchant side, it's like the exact same experience. It's still the same app for the customer that you can access everything. And then for the merchant, it's just one integration. You already integrated DoorDash. All of a sudden you get not just Dashers, but you get drones, you get autonomy, you know, you get access to all the, you know, AI tools and products that we want to ship. And I think, again, it's like, I think that's, that is like, that is like what ultimately like DoorDash is building is like, it's really like that ecosystem, uh, for local commerce. And I think that is, again, that is like something that is really hard to replicate. And I think, and I think it's, again, trying to, trying to do that in the real world across, you know, you know, like 40, 50+ countries and all these different jobs, all these different merchants. That's, that's the hard part about, about the business.

Asking for a friend, question of how you got here. Um, there is, uh, uh, an insufficient supply of researchers and people who, you know, know how to work on robotics or applied AI, uh, in the ecosystem for the recognition of all the different cool use cases you go after. Um, and a lot of people gravitate toward like the general case.

Mm-hmm.

Like, we can solve it once. Um, uh, I assume you're competing for some of those people. How do you convince people to work at DoorDash on these problems?

Yeah, my pitch is really simple. It's, it's, it's basically like, do you want to go work on prototypes and demos and, and do and be at a PhD lab? Or do you want to work on something where you can actually ship something in the real world. And I think that's kind of— yeah, I think, I think, I think that's kind of really been the culture we kind of set up, you know, both at DoorDash Labs and all the AI efforts is, is like, this is— we're not just here to do pure research. Like, at the end of the day, like, we get to ship something where we have real impact. And I think people, especially in the autonomy world for the past 10 years, were just fed up just working on something for 10 years and, you know, never actually getting to a point where they actually saw their products being used in the real world. And I think like for us, like it's like because we, like we've always been much more focused on creating, kind of taking this much more pragmatic, practical approach. Like we're not here necessarily to do like the— it's not about, oh, let's go work on like a crazy moonshot idea. It's like, let's get something out that can be shipped in the real world and actually start learning how these technologies interact with the physical world and start iterating. Because again, like technology, these things aren't built in a vacuum. You have to put something out in the real world, make contact with the real world, and actually learn, learn from that. And I think that's— we've, we did that pretty early on for, for DoorDash Dot. Actually, again, I don't think a lot of people know we've actually been doing autonomous deliveries in Phoenix for over 2 years now. We publicly announced last year or that we've been doing that for over 2 years. But really at the beginning it was just learning like, okay, like again, like I think you mentioned earlier, it's one thing to just do a fancy demo or have something that works in a one-off environment. It's entirely different to now. Okay, how do you turn this into a, an actual scaled fleet, a scaled service, a scaled business? I mean, the thing I always mention, talk about a lot is You know, building autonomy business takes more than just autonomy. It's like, how do you actually scale something in the real world, scale fleets? All of a sudden you're running into all these edge cases, right? Like you just don't see it. And when you have to do something 7, 7 days a week or 10 hours a day, 7 days a week at scale, things start breaking. I like it to be something as simple as, I don't know, like, like like a dirt covering one of your camera sensors. Okay. Like, how does— how robust is your autonomy stack able to, able to handle that? Like, there's some, there's some leaves on the ground, but it only covers kind of because again, our Dot drives on the road, but it would— it tries to act like a bike. So it will take the kind of the right side of the road or the bike lane. And if there's kind of leaves located along the kind of where the right where the sidewalks are, maybe half your wheels, the right 2 wheels are on the leaves, the left 2 wheels are still on the asphalt.

Yeah.

Well, all of a sudden the, the torque you have to send to the wheels is like very different and your autonomy stack and your, and your kind of your, um, kind of your middleware and your, and your kind of your, your kind of low-level controls has to handle that differently. Like, like that's something I would've never thought of if it was just like driving in a nice little demo environment. Uh, it's like, like things just start breaking. Like how do you handle operations? Like people don't think about actually actually, in order to scale autonomy, there's a lot of non-autonomy or like operations. Like, you have to set up depots. Again, it's a physical world business. You have to set up depots, maintenance. Like, what if your battery— like, how do you recharge your battery? Like, what if one of your braking system kind of, uh, like, over— uh, you, you have to kind of like— here, here was an issue we ran into. It's like, it's like there's certain situations where the, the vehicle has to brake so hard that it kind of the regen braking system overpowers kind of the battery because it causes this electric shock, right? Again, like it only happens like extreme edge cases, but, but there are certain situations where you have to do that because it's something in the real world, like, like safety, like this thing, like safety is like something, like it's something that's super important. So if it can't handle that, like you got to, you got to, you got to figure that out. Another example we didn't think about is, is booting up the robots. Like, like when we're doing, when this was still a demo project, we like no one thought about, oh, bootup time. All right. So, so it's literally the, the, the original version of, of, of the robot bootup was a kind of this simple Jenkins script that one of our engineers hacked together in like, in like, in like couple hours. And then, which worked, which worked fine. But then now you're doing like hundreds of robots a day, every morning needs to get booted up and the script, you know, like crashes half the time. It takes like 30, 45 minutes, but to multiply across 500 robots, all of a sudden it's like, holy crap. It's like, like there's this huge productivity— it becomes this huge productivity issue. And then, and then of course it's like, how do you think through like reliability? You know, now you have to start thinking about manufacturing, supply chain, and of course the kind of the operational aspect of actually how does, how does this thing integrate with merchants? How do you handle that? How do you do the pickup drop-off problem? How do you educate the merchant? Like, how do you even find the pin, the location of a customer's home? Which again sounds kind of silly, but when you punch in someone's address on Google Maps, like the GPS pin, it's like, especially if you're going to apartment complex, it's never kind of— I mean, I mean, it's not like always the exact same spot. Absolutely. But if you're a human, it's like you kind of figure it out, right? Like you kind of don't think about it. It's like, oh yeah, a human Dasher shows up, they can kind of find where the restaurant is.

Yeah, it's this building.

It's the building, it's at the front door. You can't do that with a robot. The robot's gonna show up to a pin and all of a sudden it's like, well, okay, which, Which, where, where, where is, which, which front, which storefront is it? Which front door is it? Which gate is it?

Now I'm just imagining Dot looking around.

Exactly.

Right.

And again, like that's something you have to figure out, but the nice thing is, again, DoorDash has that data. Like we—

Yeah.

All the drop-offs. Yeah. Like we, we can see where people are actually dropping off the package.

Yeah. Where did the human Dasher drop it off historically? And that is, you know, like, again, it's that first and last 100 feet problem. Like you don't, that, that data doesn't exist anywhere else. It doesn't exist in Google Maps.

Yeah.

It only exists. At, on, at DoorDash.

Yeah, I think that is a, uh, a really interesting and genuine advantage. Um, early on when people were like talking about what's gonna happen with AI and incumbents and startups, there were a lot of people I think had a very surface-level view of like what the incumbent data advantage was.

Mm-hmm.

Yes. Um, because they didn't like really think about like, well, what are we trying to do? Right? What is the use case? What is the intelligence supposed to accomplish? And so they'd be like, ah, like we have the I don't know, customer records in the database. And I was like, that actually has like very little to do with the thing we're trying to— we could try to accomplish with an agent, right? And I think this is totally like real in, um, in robotics where, um, I'm an investor in a company called Sunday, right? And, um, one thing that we like deeply believe in this company is you, you can't imagine the distribution, right? As soon as you like make contact with the physical world, as you said, or the like the real world, you're like, Man, if we're trying to do the dishes, why is a cat in the dishwasher? And like, you know, you're in somebody's real house and they're like, the cat likes the dishwasher.

Yeah.

And you're like, that's not, you know, that's not something you're going to go imagine. Just like, you're not going to imagine like, oh, I'm going to deal with this torque problem where like one wheel is on the leaves and not. And then you like think like, okay, but like, how important is that in the distribution? Then you find another cat in another dishwasher when you have enough data and you're like, Like, I don't know how many of these are out there, but like the only way to find out is not by an engineer sitting and being like, let me imagine this, the setup and the scenario for this robot. Like, that's clearly not going to be the reality.

I just feel like for the next frontier of AI, it's, you know, at least what we're really excited about is like what, how it's going to affect the physical world, you know? And I think to your point, it's like you can only simulate so much, you can only like, you know, uh, you know, pretend and imagine various demo situations. So, um, I think one thing that we are very— I think another thing that makes us very confident is like pairing that world-class operational expertise that we have with world-class technology.

Mm-hmm.

And I think, you know, a lot of AI researchers are very hesitant to do a lot of the operational stuff, or they think it's like easy to handle. But I think one thing that's really powerful about what we have here at DoorDash is we have a world-class operations team that you can partner with, whether it's to collect or annotate data, whether it's to figure out how to deploy robots and figure out how to, like, get the fleet operations to work. And I think for a lot of people we talk to, that's very compelling because it's like, hey, actually there's a— we're not just talking hypothetical here, you know.

You're making the deliveries in Phoenix. What are the challenges from here for scale up?

I mean, we've been doing delivery and things for over 2 years now. I mean, we went fully autonomous L4 last year. I mean, it's like we— I think that was a super exciting milestone. And really it's just a matter of like, how do you take this from— again, it's like originally it was just a couple of robots, 10 robots to 100. Again, it's just like we got to make that hill climb of like, how do you, how do you scale this? And I think it's really 3 components is can we get the autonomy to scale? 5 years ago, the question was like, was autonomy even possible? Like, was this, was this just a research project? Is this a science fiction? Um, you've seen kind of now with, especially with AI, like Waymo's kind of made that breakthrough. I think Tesla's starting to make that breakthrough. We made that breakthrough last year. Um, like our entire autonomy stack is built in-house, but purpose-built for, for delivery, which is, again, it's, it's a little bit different. You can't, it's not just copy. I think this is the, the other thing people miss is you don't, you can't just copy and paste what Waymo's done. And then plop it into the DoorDash Dot and everything works. It's, again, it's the use case is a little bit different. This is a bike claim profile vehicle, but it's constantly navigating between the road and the sidewalks. As far as I know, this is like— there's nothing else like this in the world besides that even behaves like DoorDash Dot. But we've kind of built it because we kind of built it uniquely to our use case. So autonomy is definitely one piece, like how to keep scaling, um, across not just Phoenix, but we want to bring to Bay Area, more cities, you know, that I'm sure we're going to run into more, more and more edge cases. Um, but the funny thing is like the autonomy is probably increasingly becoming less and less of a constraint, of a blocker. It's really like now how do you— it's really more the next two, which is Um, the second is like operational, like how do you scale operations? Restaurants behave in, in Phoenix look different than restaurants in, in San Francisco versus like, you know, London versus Helsinki. How do you adapt to all these different integrations? How do you—

So it's the interface layer and then like the fleet management of it?

Yeah, interface and fleet management. And then the last piece is, is, is hardware. Like how do you— and it's kind of funny, it's like a when we first started like 5 years ago, like everyone thought hardware was a commodity and now it's starting to look like hardware is starting to become a bottleneck. Like it's like we hand-built the first 100 robots ourselves and which is not an issue, but then okay, the next 1,000 or 10,000, well, we're going to have to now start to think of things like supply chain and like, like component reliability. Like it's, it's like, it's like these things have to last for a really long time. It's like, how do you think about Yeah, it's like it's—

and you're not guessing because you can actually tell how long it needs to last and how it's doing the field.

Exactly right. Like manufacturing, you know, like it's like, it's like learning, learning all that. And that turns out to be a pretty hard problem at scale. And so, so one of the things we actually did is we actually partnered up with this company called ALSO, which is this micro-mobility company that spun out of Rivian. So, um, RJ is actually the board founder and chairman of the company. So if, you know, like, why don't we work with someone who knows how to actually scale vehicles? And, and so, so that's kind of one of the, uh, partnerships we, we struck up. But it's kind of funny, it's like the problem 5 years ago was autonomy. Now it's increasingly becoming more about operations, commercialization, hardware manufacturing. And again, it's like this is— I feel like this is where DoorDash gets to shine with our scale advantage and operation advantages. How do we take this thing from not just 0 to 1, but like 1 to 100, 1 to 1,000, 1 to 3 billion, 1 to 3 billion. And I feel like DoorDash is just so well positioned to take on this. It's like we have just such a unique advantage here. And I think that's, that's what, that's where we want to play in terms of our, play to our strengths.

So you have these enormous strengths, you've got the network and the existing great business and these like two, you know, amongst others, I'm sure, like two really big plays around agentic commerce and around autonomy. How do you think about just, it's, it's a 10,000+ person company and like a lot of that company is ops, a lot of that company is technology. Um, uh, I'm sure you're thinking deeply about productivity of that workforce, like who owns it, what matters today. You're even publishing benchmarks. Like, talk about that.

I feel like in the past couple of years, what was required to really operate at a high level in the technology industry has changed a lot. And I think one of the reasons why we were so excited to acquire a company called Metis last year was really to just infuse some of that AI-native thinking into the company. And I think for a company of our size, it's been really And I think every company is— every large company at least is facing it. I think a lot of startups, I mean, you see this better than anyone else probably, is like the way they operate is so different. And I think a lot of people at our company, they, they have struggled to see what's possible because they're so used to how things have worked historically. And so I think really figuring out how do we bring in people who actually have seen what is possible on the frontier and incorporating that into how we do our work. And I think, you know, coding is obviously like the most obvious place to do transformation, and we've seen a lot of gains there. But there's also work we're doing in terms of how do we do AI enablement across the entire organization. And so I think, you know, figuring out how to benchmark various parts of the company. I think we announced a benchmark called Dashbench a couple weeks ago now that was mainly focused on our ability to figure out how well various models and Harness performed on coding tasks. And so that was a really good initial exercise for us to figure out how do we calculate the ROI on all this money we're spending. I mean, I think I was looking at it a week ago. I think our spend in June went up like 20x versus what the spend was in January.

Wow.

Yeah. And so I think it's like, okay, like clearly this has got to get some sort of return. And so, and obviously, like, I think we're seeing a lot of, you know, suggestion.

Wait, can I ask? You can, you can, uh, not answer, but like, since you have inspected this spend, like, has it come down? Has it been flat? Has it continued to grow?

Um, we're seeing a flatline. Okay. Um, and I think a lot of it is through some of these intentional efforts. Like, cause I think, you know, when people were experimenting with, especially at the beginning of the year, or like maybe like December last year, it's like, I think there was just a step function change in terms of what was possible. I think a lot of it was just experimenting and letting people run with it. But it's gone to a point where it's like, okay, one, there's easy things we can do to make sure that we're not doing wasteful stuff. But two is, as it relates to this benchmark that we released, it's like, okay, we actually need to start calculating the ROI. If there's a way for us to maximize the intelligence, but maybe delegate to open-weight models for some of the cheaper tasks, we can actually do— we can get the fabled level of intelligence but actually pay less than if we were just using these closed-weight models. So I think coding is kind of where we think there's a lot of opportunity, mainly because, I mean, the vast majority of that spend is still within like engineering-related tasks. But we're actually seeing the highest amount of growth in our organization in terms of like seats in the non-technical organizations because, you know, analysts are finding a lot of value in it. Our operators, you know, account managers who are trying to figure out, okay, how do I do my VR with the strategic merchants? How do we, like, automate a lot of that? And so I think, you know, there's work we're doing there to figure out, okay, how do we benchmark some of the work we're doing in some of these other areas? And I think another thing that is interesting for us is because we work with some of these frontier labs on like, okay, like for like accounting tasks or analytics tasks, like how well do the latest models perform?. And I think a challenge that we've run into is like, we'll ask our teams like, hey, how well do the models perform on your task? They're like, you know, it works okay. And I think, you know, but then when we do—

and you're like, okay, like $30 million of okay.

Yeah, exactly. It's like the, the cost. But then it's like, okay, when we, then when we send some of these data to the labs, we'll have to do like the data scrubbing and then we'll have to like, you know, you know, put in like RL environment, whatever. And then, you know, then the models crush it. But then we're like, there's clearly— it's kind of like what you're saying with like the Sunday Robotics example. It's like, okay, if you like dumb down the problem, maybe the models do well. But like for some reason, and when we actually have it with the enterprise data and all the real stuff, it's not performing as well. And so I think for us, it's a question of like, hey, is it because like there's just things that we need to do with the harness to get the model to perform better? Or are there inherently things that the models just don't have in their data distribution or whatever capability set that is not allowing that step function change enablement in like accounting analytics or, you know, finance functions. And so I think that's like kind of like the next step for us beyond the coding stuff, which of course is a lot of work for us to do. But I think there's a lot of interesting things in terms of like how do we really see that step function change across the work.

Is the long-term view like you get rid of all the Dashers and it's just Dots everywhere? What happens?

Yeah, well, my take, my prediction actually is in a world where robotics, drones, AI is, is everywhere, uh, my guess is that in 10 years' time, we're actually gonna have more Dashers doing, doing deliveries, not less. Uh, simply just because, again, I, I think it's just the Well, one, I think the pace at which DoorDash is growing is just— I mean, and then the scale at which we're operating is pretty insane. I don't know if people know, but like we have over 9 million Dashers doing deliveries and the business growing 25% year over year. Like fast forward 10 years' time, like, like, and if we want to 5x from here, 10x from here, well, where are they? Where's the supply going to come from? Like, are you going to have half America doing, doing deliveries for us every month? Like, that's probably not going to be the case. Like, like there has to be— we're going to have to find other areas of opportunity to both bring in new modalities as well as improve efficiencies within our business. And I think, and I think DOT robotics, drones like Waymo's, like sidewalk robots, I think we're going to— you're going to see a world where we're going to have this multimodal fleet, like we're going to need our hand, get our hands on every single modality we can get. So I think you're not only going to see more autonomy and more robotics, but I think you're going to see even more humans as well. And I mean, I mean, and, and I think, and I also just think like with the introduction of autonomy and robotics, like, and efficiency gains you're going to see over time, like I also think you're just going to see an even stronger surge in demand as autonomy, as delivery becomes even more affordable.

And I look forward to getting 60s a day.

Yeah.

Amazing. And Andy, when you think about what you've learned with the initial forays into agent commerce, like how are people going to buy differently in the future beyond food?

Yeah, I mean, I think one of the trends that I found fascinating is like over the past couple of years, Google search query lengths have gone longer., and I think to me how I've translated that is like, okay, people feel more comfortable like talking to like agents or to like apps like they would a normal human being. And so I think if we fast forward and look ahead to the future, I think the easier we can make it for people to kind of interface with apps or with agents like they would with a person, I think it's going to reduce the friction in terms of compelling them to place an order, whether that's for food or for like their groceries or for retail, what have you. And I think another thing that I think is going to be true is I think we're all going to need to think about like, what does the agent-first experience look like? And, you know, I think we've been testing some of that with the recent DoorDash CLI that we launched last week. But I just think there's a lot of interesting emerging use cases that can crop up, once, once you start thinking about this. Like, one concrete example I can talk about is like someone who was really excited to use the DoorDash CLI because they're like, hey, let me like basically streamline my office manager use case for my startup. And when they found out that DoorDash did more than just lunch, they're like, oh, actually, wait, DoorDash can order me like convenience and groceries. So then they just pointed a camera at their pantry shelf, and whenever the shelf was getting empty, like they would fire off the agent to basically restock the shelf. So I think those types of use cases that you wouldn't really think of, but I think it's going to unlock some interesting use cases that I think would not really be as feasible or possible like in today's world. But as we make things more naturally agent-first, I think some of these use cases are going to become a lot more interesting.

Amazing. I love how ambitious you guys are for both the user experience and the, Scope and Scale at DoorDash.

Thanks, guys.

Yeah, it's a pleasure to be here.

Find us on Twitter @NoPriorsPod. Subscribe to our YouTube channel if you want to see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way you get a new episode every week. And sign up for emails or find transcripts for every episode at no-priors.com.