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 and Stanley. Thank you so much for being here. I’m 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. What was the backstory here?

It started a couple of years ago, honestly, in terms of our attempts to try to make a play here. We were originally bullish on voice as the modality. That ended up not being the thing, but maybe it will be in the future. It just didn’t really land.

The thing that was very interesting for us was this natural, conversational experience. What we’ve seen is that people can naturally translate what’s in their head into this interface, versus trying to do research online or some kind of keyword optimization. People found it easier to search for things, whether more nuanced restaurant-discovery searches or different tasks on the grocery side. We’ve seen a lot of interesting traction that has held up as we’ve expanded the rollout.

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

On the restaurant side, we’re seeing that 50% of the trajectories of people using Ask DoorDash for restaurants involve ordering from places they’ve never ordered from before. That’s huge, because that’s one of the hardest metrics historically for DoorDash for us to move.

On the grocery side, we’re seeing much higher basket sizes—about 40% larger, I would say. People will take a picture of what’s in their fridge and say, “Help me stock up my fridge,” or they’ll do meal planning when they have dietary constraints. They might say, “Hey, I want to cook a pasta dinner this weekend with my family.” Or they’ll say, “Help me reorder my usuals,” and 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 that suggests there’s actually latent demand that wasn’t being served because it wasn’t easy enough to eat at new places.

Correct. I think a lot of people on the restaurant side build habits, but people also want some diversity in terms of what they’re eating. We felt like this experience ended up being a natural way to allow people to express that.

Think about the social currency of, “My friend Andy found a really good new restaurant for me. Andy’s awesome,” right? I feel like that’s even a different way people look at DoorDash.

Another investment we made was incorporating world knowledge into the experience.

What does that mean here?

Things that are going on with restaurants outside of DoorDash. We’ll see what’s trending on the internet or what people are talking about in various forums—things that aren’t in the models because their knowledge cutoff is too early. Maybe people want to know what’s trending online or what’s cool to eat, so that was something we tried to incorporate into the experience to make people trust it more.

How do you think people will buy from or think about restaurants differently five years from now?

I don’t know about five years from now.

I realize it’s really hard in the age of AI to think about the next step.

For Ask DoorDash, to start with, maybe the next couple of months, I think it’s about making it easier for people to discover the experience and figure out what to do. It can be intimidating if you just see suggested queries that you can type, because some people don’t know where to start. We’re figuring out how to experiment and tinker with the user experience to encourage people to find use cases for it.

If I think further out, it’s a little more speculative, but Stanley and I talk about this all the time. If someone were to create DoorDash today—I don’t know, college kids in a garage trying to start DoorDash—I think it would look very different, probably more agentic-first. One statistic that I always like to think about nowadays is that there’s more agent traffic on the web than human traffic. How do we have a DoorDash-type experience that plays into that trend? I think there are some interesting speculations there, but it’s hard to say.

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

One cool example is someone saying, “For our office, I can have one of the cameras on the pantry shelf. When the shelf starts to get empty, I can fire off a query to DoorDash to stock up my shelf.”

Yes, this is a human-being task here. Yes.

Yeah. That was kind of an example of our early experimentation with our CLI. It’s about making it less friction for an agent to participate in that experience.

While we’re here talking about user needs, I’ve got to be a top-percentile DoorDash consumer. I host family dinner for extended family every Sunday night, and we eat DoorDash because I’m going to cook for all these people every week. I can’t do it all the time. I do the same thing every time: I poll everyone—“Okay, who’s coming?”—and then these people have allergies and whatever else. I ask, “Does anybody feel like anything special?”

Oh, yeah.

Then I order, and I feel like that’s all within the realm of possibility. You just put it on autopilot for me. I show up and my family’s good.

That is a use case that is—not exactly the same, but a similar use case is office lunch ordering. If you’re the office manager, you don’t want to have to remember to order lunch at a certain time, because otherwise it’s not going to show up. Everyone has their own allergies, dietary preferences, and so on.

Stanley, you guys are doing a whole bunch of things on the autonomy and robotics side as well. Your view of DoorDash as founders is clearly broader and more ambitious than the surface-level view of it as a food-delivery network, or whatever the first one-liner for the company was. How long ago did the robotics efforts start?

We’ve actually been looking into robotics and autonomy much longer than people thought—since 2018, actually, back when it wasn’t obvious that autonomy and robotics were going to be a thing. We felt like this was going to be a technology that was transformative to our space and potentially disruptive.

I think that’s the nice thing about being a founder-led company: We get to think about much more future-speculative things that are on the horizon and constantly think about how we make sure we don’t get disrupted by the next—

I think, like Andy said, the next DoorDash, if it comes along, is not going to be someone that builds the exact same version of DoorDash, but with a better UI.

That would be dumb.

Yeah. It’s going to be something like, “Okay, how do we incorporate AI agent commerce? How do we incorporate autonomy, robotics, drone deliveries?” and so on.

Fast-forward 7 or 8 years, and I think you’re seeing everything starting to play out in AI, robotics, and autonomy. You’re seeing way more stuff happening, and I think we’re glad that we made that investment early, in 2018.

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

I think it’s probably the same way we invest in a lot of things at DoorDash: Everything starts out as experiments. In a way, that was the founding story behind DoorDash. DoorDash was a Stanford college dorm-room experiment.

It started out as a website called politely.com, with 8 PDF menus and a Google Voice phone number. It was only once we figured out, “Okay, there’s something here. Let’s turn this into a company.” That’s basically the philosophy we’ve taken throughout the past 13 years, and we’ve applied it to autonomy and AI as well.

When we first started in 2018, the intention wasn’t, “Hey, let’s go spin up this giant robotics program. Let’s hire a roboticist and go build hardware.” It was really me and half an engineer’s time. It was a skunkworks project, an experiment to explore what was out there. We didn’t even know what autonomy looked like or how robotics was going to impact our space, but we wanted to explore, form partnerships, learn, and experiment.

In the beginning, the intention wasn’t to build our own robot. We didn’t think we needed to build any of this technology. We thought, “Okay, we can just partner up with a bunch of folks.” Back then, we didn’t know anything about robotics. There were all these startups out there that had built robots and autonomy, so why don’t we just work with them? We could essentially just be the platform: we’d build the APIs, handle the distribution, and so on.

We did that for several years. We worked with everyone in the space, from the sidewalk robot players all the way up to the robo-taxi players. I’d say there are 3 things we learned through that experience. The first is that it validated or confirmed our belief that there’s something here: autonomy is a question of when it was going to happen, not if. Fast-forward to today, you’re seeing the Waymos driving, right? It’s happening, so we should keep investing.

The second is that it allowed us to learn what it takes to actually enable autonomy. It turns out there are a lot of things you have to build around autonomy: the infrastructure, the ecosystem, how autonomy integrates with DoorDash, what deliveries you take on, and the operational aspects. There are a lot of things you have to build around autonomy in order to make autonomy possible. It’s not just that you plop a robot in, or even plop an LLM in, and then things magically happen. There are a lot of things around it, and you 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 asks: What are all the products and technology—the APIs and the dispatch—you need to build now that we’re in a post-autonomy world, where autonomy, 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?

The last thing, which I think is probably the most important thing we learned, and which eventually led us to realize we had to build this technology ourselves, is really this idea of building toward a use case.

Mhm.

Yes. There's a lot of autonomy startups out there. But it always felt like these companies weren't really focused on a use case. It always felt like they kind of built the technology first.

Mhm.

And then retroactively tried to go find a problem to fit into. These things were all built in a vacuum, which is kind of weird, because in the software world, when we went through YC, we were always taught to serve the customer and build something people want. That's drilled into you, and then you can iterate. But when it comes to hardware and hard tech and AI and robotics, people just do the opposite, where they try to build the tech first and not really think about the use case they're building toward. Whenever that happens, you end up with something that just wasn't quite the right fit. We went through this process where a lot of these companies were out there, but it always felt like they weren't exactly what DoorDash needed.

For example, in the autonomy world, there are basically two buckets of companies. You have these sidewalk robot companies, which are two- or three-mile-per-hour water coolers on wheels—super effective, simple technology. But we quickly realized that speed and distance were huge limitations, because the average delivery at DoorDash is about 3 to 5 miles, and the typical delivery time is about 15 minutes, excluding the time it takes to make the food. If you put a 2-mile-per-hour sidewalk robot on that, it's just never going to work.

On the other end of the spectrum, you have the robo-taxi players, which are designed for carrying people around. It's a 4,000-pound vehicle that goes super fast. You're transporting people, but the problem of carrying people and carrying goods is actually a little different. If you're only carrying a couple of burritos around, do you really need a 4,000-pound car with chairs and air conditioning? The pickup and drop-off problem is also very different in robo-taxis. You can walk to a Waymo. 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? That's totally fine because you can walk, but packages can't do that. That's 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 on the customer side, how do you drop off the food? How do you find the driveway? People expect their food to be dropped off, or the vehicle to be pulled up straight to the front of their driveway or porch.

When we looked around and asked ourselves, if you were to start from first principles, from the business and the customer use case, and work your way backward, what would that look like? It turns out no one was really building that. It's not a sidewalk robot, and it's not a robo-taxi. We felt like it was probably something in between. If you're trying to solve that 3-to-5-mile delivery in a dense suburb, which is where most of the deliveries happen, the right metaphor is probably an autonomous motorcycle, scooter, or bike-profile vehicle. It doesn't need to be 4,000 pounds. It's probably 300 pounds, but it also has to be a lot faster than a sidewalk robot. Let's go 20 or 25 miles per hour.

When we looked around and saw no one was building that, we decided that instead of waiting around, we were going to control our own destiny. Let's invest in this and see what we can build. It took many iterations. We started testing this with real DoorDash deliveries, looking at the 10 billion deliveries we've done, extracting the insights and operational learnings we had, and that's eventually what led us to launch and ship our in-house autonomous delivery robot. It's been quite a journey, but this is something we look to bring to every aspect of the business, whether it's autonomy, robotics, or AI. It always starts out as an experiment. It always starts out with: What is the customer problem you're solving for? What's the use case you're solving for? Work your way backward, then iterate and validate 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. I have a hypothesis, and I’m curious if it resonates with either of you about why a lot of people in this era are building technology first instead of starting with the customer. I think people think everything is going to work like ChatGPT.

Mhm.

Right. By the way, there was, of course, work done on instruction fine-tuning to get it shaped into a product that was still a user experience. But I think the mental model people have—that it’s a general technology and it’s just free to turn into different applications—is what they’re applying to lots of different things now, especially in autonomy. My sense is people are like, “Okay, we’ll make the model, and then the other stuff will be, if not easy, at least secondary.” This is not my view at all.

That’s basically your methodology for building the Dot form factor.

Yeah, I agree with you there. I think maybe that approach works in software land, but for a business like ours, DoorDash is a physical-world business. You bring technology into the physical world, and the physical world is always a lot messier. It’s a lot more complicated and a lot more nuanced.

I think one of the things people don’t realize is just how complicated DoorDash is. We do over 3 billion deliveries a year. No 2 deliveries look the same. All 3 billion deliveries look different, and they come in all sorts of shapes and sizes and different geographies. A delivery in downtown San Francisco is completely different from a delivery in Dallas, or even in Europe or Helsinki, where it’s snowing. A pizza is very different from ice cream. Your dinner is very different from your grocery order, which is very different now that we’re expanding into retail, 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 this problem is. And that's kind of what we have to solve for at DoorDash. I think that's part of the learning process, especially when it comes to building autonomy or even AI: How do you manage through all that complexity? Again, it always comes down to: Do you understand the use case? 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—over 40 million consumers ordering every single month—and we understand the complexities of how to handle when things go wrong and how to integrate across all different types of merchants. The way you work with a McDonald's or Starbucks is very different from working with a mom-and-pop sandwich shop. A drive-thru restaurant is very different from a restaurant at a strip mall or on downtown Main Street.

How do you handle those different use cases? Different interactions, different pickup points. I don't know if there's anything you want to add on the AI side. For me, that analogy you brought up, I think about it in terms of the autonomy thing, but I also think about it in terms of how the humanoid robotics space is starting to play out, potentially.

We also launched a product called Tasks a couple of months ago, where we're having people in the Dasher fleet help collect data points to help train some of these world models. I think we're so early there, and there are so many different form factors that you can use. There are also different opinions on what type of model is going to work versus not.

But unlike something like ChatGPT, I think there's a lot of expense needed just to invest in the V1 of this. I guess ChatGPT could cost a lot of money, too. But I think there's a lot of pressure to figure out how I actually provide value. I have to be better than what people can do today. Whether it's Dot delivering something end-to-end, or—you probably invest in a bunch of different players in the space—but there's real pressure to be better than the alternative, either from a quality or a cost perspective.

Yes. Otherwise, what are we doing?

Yeah, exactly.

So, for those of us who aren't in Phoenix, what is DoorDash Dot, and tell us about the design of it?

DoorDash Dot is an autonomous delivery robot. It's built entirely in-house at DoorDash. It weighs 300 pounds, travels up to 20 miles per hour, and is one-tenth the size of a car. It's the only delivery robot out there that's designed to travel not just on sidewalks, but also in bike lanes and on the road.

It's live in Phoenix. We've been doing deliveries for almost 2 years now. It's fully autonomous, Level 4. If you come up to Phoenix, to Tempe, it really feels like Waymo in San Francisco.

Elad Gil

I'm going to state something and see if this is correct, or if you agree: Even beyond understanding the wealth of use cases, 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. Anyone familiar with the area understands that it's not that hard to get a cherry-picked demo of 1 cool success on a task, right?

The problem is getting it to work on any object or in any environment. So there's this huge question in the industry: How are we going to get data that feels like realistic data? The best realistic data is actually real-world data. I think that's a really interesting premise for why you might have the right to go do this, besides wanting to do it for the quality of your business.

Yeah, no, exactly. I think that's also where DoorDash gets to shine with our advantage. We don't necessarily have to solve for 100% of our use cases. That was also part of the learning in our early years, when we did the partnerships for how we built our autonomous delivery platform: understanding what kinds of deliveries fit into what modality.

I think the vision was always, let's not design something to solve for everything. Instead, let's go with a multimodal strategy, where perhaps you have DoorDash Dot to do the 3-to-5-mile suburban deliveries from a strip mall. Right now, we're live in Phoenix. That's kind of the perfect market for Dot: dense suburbs, yet things are still far enough apart.

If it's a rural area where there's poor road infrastructure, maybe you send a drone delivery for that. If it's a lightweight order, maybe you send a drone delivery. If it's a complicated, multistep grocery order where we have to go up and down stairs and pick and pack orders, you're still going to have a Dasher for that.

I think that's the nice thing about DoorDash: You don't have to take an all-or-nothing approach. You can phase in these modalities over time and pick and choose the right one. 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? Are there certain deliveries we can carve out that make 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 accomplish this task.”

Exactly. And then, from the consumer side and the merchant side, it's the exact same experience: still the same app for the customer, where you can access everything. And then for the merchant, it's just 1 integration.

You've already integrated DoorDash. All of a sudden, you get not just Dashers, but drones and autonomy. You get access to all the AI tools and products that we ship. I think that's ultimately what DoorDash is building: It's really that ecosystem for local commerce.

I think that's something that's really hard to replicate. Trying to do that in the real world across 40, 50-plus countries and all these different geographies, all these different merchants—that's the hard part about the business.

Elad Gil

Asking for a friend: A question about how you got here. There's an insufficient supply of researchers and people who know how to work on robotics or applied AI in the ecosystem, given all the different cool use cases you go after. A lot of people gravitate toward the general case, like, “We can solve it once.”

I assume you're competing for some of those people. How do you convince people to work at DoorDash on these problems?

My pitch is really simple. It's basically: Do you want to go work on prototypes and demos and be at a PhD lab, or do you want to work on something where you can actually ship something in the real world?

I think that's the culture we set up, both at DoorDash Labs and across all the AI efforts. We're not just here to do pure research. At the end of the day, we get to ship something where we have real impact.

I think people, at least in the autonomy world for the past 10 years, were just fed up with working on something for 10 years and never actually getting to a point where they saw their products being used in the real world. For us, we've always been much more focused on taking this much more pragmatic, practical approach.

We're not here necessarily to do a crazy moonshot idea. It's, “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.” 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 from that.

I think we did that pretty early on at DoorDash. I don't think a lot of people know that we've actually been doing autonomous deliveries in Phoenix for over 2 years now. We publicly announced last year that we'd been doing it for over 2 years, but in the beginning it was just learning. As you mentioned earlier, it's one thing to do a fancy demo or have something that works in a one-off environment. It's entirely different to ask: How do you turn this into an actual scaled fleet, a scaled service, a scaled business?

The thing I always talk about is that building an autonomy business takes more than just autonomy. How do you actually scale something in the real world? How do you scale fleets? All of a sudden, you're running into all these edge cases, right? You just don't see them.

When you have to do something 7 days a week or 10 hours a day, 7 days a week at scale, things start breaking, right? It could be something as simple as dirt covering one of your camera sensors. How robust is your autonomy stack able to handle that? There could be some leaves on the ground, but it only covers—because, again, our Dot drives on the road, but it tries to act like a bike. So it’ll take the right side of the road or the bike lane.

If there are leaves along where the sidewalks are, maybe the right 2 wheels are on the leaves and the left 2 wheels are still on the asphalt.

Yeah.

Well, all of a sudden, the torque you have to send to the wheels is very different, and your autonomy stack, middleware, and low-level controls have to handle that differently. That’s something I would have never thought of if it were just driving in a nice little demo environment. Things just start breaking. How do you handle operations? People don’t think about it, but in order to scale autonomy, there’s a lot of non-autonomy, or operations.

You have to set up depots and maintenance. It’s a physical-world business. How do you recharge your battery? Here’s an issue we ran into: there are certain situations where the vehicle has to brake so hard that the regenerative braking system overpowers the battery because it causes an electric shock, right? Again, it only happens in extreme edge cases, but there are certain situations where you have to do that because it’s something in the real world. Safety is super important, so if it can’t handle that, you’ve got to figure that out.

Another example we didn’t think about is booting up the robots. When we were doing this as a demo project, no one thought about boot-up time. The original version of the robot boot-up was a simple Jenkins script that one of our engineers hacked together in a couple of hours, which worked fine. But now you’re doing hundreds of robots a day. Every morning, each one needs to get booted up, and the script crashes half the time. It takes 30–45 minutes, but when you multiply that across 500 robots, all of a sudden it becomes a huge productivity issue.

Then, of course, there’s reliability. Now you have to start thinking about manufacturing and the supply chain, and, of course, the operational aspect of how this thing interfaces with merchants. How do you handle the pickup and drop-off problem? How do you educate the merchant? How do you even find the pin location of a customer’s home?

That sounds silly, but when you punch someone’s address into Google Maps, the GPS pin—especially if you’re going to an apartment complex—is not always in the exact same spot. But if you’re a human, you figure it out, right? You don’t think about it. A human Dasher shows up, and they can find where the restaurant is, the building, and the front door.

You can’t do that with a robot. The robot is going to show up to a pin, and all of a sudden it’s like, “Well, okay, which one? Where’s the storefront? Which front door is it? Which gate is it?” Now just imagine Dot looking around.

Exactly. That’s something you have to figure out. But the nice thing is that DoorDash has that data. We can see all the drop-offs—where people are actually dropping off the package.

Yeah. Where did the human Dasher drop it off historically? That’s the first-and-last-100-feet problem. That data doesn’t exist anywhere else. It doesn’t exist in Google Maps. It only exists at DoorDash.

I think that’s a really interesting and genuine advantage. Early on, when people were talking about what was going to happen with AI and incumbents and startups, there were a lot of people who had a very surface-level view of what the incumbent data advantage was. They didn’t really think about, “What are we trying to do? What is the use case? What is the intelligence supposed to accomplish?” They’d say, “We have the customer records and database,” and I was like, “That actually has very little to do with the thing we’re trying to accomplish with an agent.”

I think this is totally real in robotics. I’m an investor in a company called Sunday, and one thing that we deeply believe in at this company is that you can’t imagine the distribution. As soon as you make contact with the physical world, or the real world, you’re like, “Man, if we’re trying to do the dishes, why is a cat in the dishwasher?” You’re in somebody’s real house, and they’re like, “The cat likes the dishwasher.” You’re like, “That’s not something you’re going to imagine.”

Just like you’re not going to imagine, “Oh, I’m going to deal with this torque problem where 1 wheel is on the leaves.” Then you think, “Okay, but how important is that in the distribution?” You find another cat in another dishwasher when you have enough data, and you’re like, “I don’t know how many of these are out there, but the only way to find out is not by an engineer sitting there and saying, ‘Let me imagine the setup and the scenario for this robot.’” That’s clearly not going to be the reality.

I just feel like, for the next frontier of AI, at least what we’re really excited about is how it’s going to affect the physical world. To your point, you can only simulate so much. You can only pretend and imagine various demo situations.

I think one thing that makes us very confident is pairing the world-class operational expertise that we have with world-class technology. A lot of AI researchers are very hesitant to do a lot of the operational stuff, or they think it’s easy to handle. But one thing that’s really powerful about what we have here at DoorDash is a world-class operations team that you can partner with, whether it’s to collect or annotate data, to figure out how to deploy robots, or to figure out how to get fleet operations to work.

For a lot of people we talk to, that’s very compelling because it’s like, “Hey, actually, we’re not just talking hypothetically here.”

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

We’ve been doing deliveries in Phoenix for over 2 years now. We went fully autonomous Level 4 last year. That was a super exciting milestone. Really, it’s just a matter of how you take this from—originally, it was just a couple of robots, then 10 robots, then 100. We have to make that hill climb: how do you scale this?

I think there are really 3 components. Can we get the autonomy to scale? 5 years ago, the question was, was autonomy even possible? Was this just a research project? Was this science fiction? Now, especially with AI, Waymo has made that breakthrough. I think Tesla is starting to make that breakthrough. We made that breakthrough last year.

Our entire autonomy stack is built in-house and is purpose-built for delivery, which is, again, a little bit different. You can’t just copy and paste what Waymo has done, plop it into the DoorDash Dot, and have everything work. The use case is a little bit different. This is a bike-lane-profile vehicle that’s constantly navigating between the road and the sidewalks. As far as I know, there’s nothing else like this in the world that even behaves like DoorDash Dot. We built it uniquely for our use case.

Autonomy is definitely 1 piece. How do you keep scaling across not just Phoenix, but also bring it to the Bay Area and more cities? I’m sure we’re going to run into more and more edge cases. The funny thing is that autonomy is probably increasingly becoming less and less of a constraint or blocker.

The next 2 pieces are operations and hardware. How do you scale operations? Restaurants in Phoenix look different from restaurants in San Francisco, London, or Helsinki. How do you adapt to all these different integrations?

So it’s the interface layer and then the fleet management of it.

Interface and fleet management. The last piece is hardware, which is kind of funny.

When we first started 5 years ago, everyone thought hardware was a commodity, and now it's starting to look like hardware is becoming a bottleneck. We hand-built the first 100 robots ourselves, which was not an issue. But then, okay, what about the next 1,000 or 10,000? We're now starting to think about things like supply chain and component reliability. These things have to last for a really long time.

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

Exactly. Right. Manufacturing and learning all that turns out to be a pretty hard problem at scale. One of the things we actually did was partner with a company called Also, which is a micromobility company spun out of Rivian. RJ is actually the founder and chairman of the company.

Why don't we work with someone who knows how to actually scale vehicles? That's one of the partnerships we struck up. It's kind of funny: 5 years ago, the problem was autonomy. Now it's increasingly becoming more about operations, commercialization, hardware, and manufacturing.

Again, I feel like this is where DoorDash gets to shine, with our scale advantage and operational advantage. How do we take this thing from not just 0 to 1, but 100 to 1,000?

1 to 3 billion.

Yeah, 1 to 3 billion. Right. I feel like DoorDash is so well positioned to take this on. We have such a unique advantage here, and I think that's where we want to play, in terms of playing to our strengths.

You have these enormous strengths. You've got the network and the existing great business, and these, amongst others, I'm sure, 2 really big plays around agent commerce and autonomy. How do you think about it? It's a 10,000-plus-person company, and a lot of that company is ops, while a lot of that company is technology. I'm sure you're thinking deeply about productivity of that workforce—who owns it, what matters today. You're publishing benchmarks. Talk about that.

In the past couple of years, what was required to really operate at a high level in the technology industry has changed a lot. One of the reasons why we were so excited to acquire a company called Metis last year was really to infuse some of that AI-native thinking into the company.

For a company of our size, I think every large company, at least, is facing this. A lot of startups—you see this better than anyone else, probably—the way they operate is so different. A lot of people at our company have struggled to see what's possible because they're so used to how things have worked historically.

We're really figuring out how to bring in people who have actually seen what's possible on the frontier and incorporate that into how we do our work. Coding is obviously the most obvious place to do a transformation, and we've seen a lot of gains there. But we're also working on how to do AI enablement across the entire organization.

We're figuring out how to benchmark various parts of the company. We announced a benchmark called DashBench a couple of weeks ago, which was mainly focused on our ability to figure out how well various models and harnesses performed on coding tasks. That was a really good initial exercise for us to figure out how to calculate the ROI on all this money we're spending.

I was looking at it a week ago, and I think our spend in June went up 20x versus what the spend was in January.

Wow.

Yeah. Clearly, this has got to get some sort of return.

Can I ask you—can you not answer, but since you've inspected this spend, has it come down, been flat, or continued to grow?

We're seeing it flatline. Okay. A lot of it is through some of these intentional efforts. When people were experimenting, especially at the beginning of the year, or maybe in December of last year, there was a step-function change in terms of what was possible. A lot of it was just experimenting and letting people run with it, but it's gotten to a point where there are easy things we can do to make sure we're not doing wasteful stuff.

The second thing is that, as it relates to the benchmark we released, we actually need to start calculating the ROI. If there's a way for us to maximize the intelligence but delegate to open-weight models for some of the cheaper tasks, we can get frontier-level intelligence but pay less than if we were just using these closed-weight models.

Coding is where we think there's a lot of opportunity, mainly because the vast majority of that spend is still within engineering-related tasks. But we're actually seeing the highest amount of growth in our organization, in terms of seats, in the nontechnical organizations. Analysts are finding a lot of value in it, as are our operators and account managers, who are trying to figure out how to do QBRs with strategic merchants and automate a lot of that.

We're working there to figure out how to benchmark some of the work we're doing in these other areas. Another thing that's interesting for us is that we work with some of these frontier AI labs on accounting tasks and analytics tasks to understand how well the latest models perform.

We'll ask our teams, "Hey, how well do the models perform on your task?" They're like, "Yeah, it works okay." But then, when we send somebody's data to the labs, we have to do the data scrubbing and put it in an RL environment, and then the models crush it.

It's kind of like what you're saying with the Sunday Robotics example: if you dumb down the problem, the models do well. But when we actually have enterprise data with all the real stuff, it's not performing as well.

For us, the question is: Is it because there are things we need to do with the harness to get the model to perform, or are there inherently things that the models just don't have in their data distribution or whatever capability set that are not allowing that step-function change in enablement in accounting, analytics, or finance functions?

That's the next step for us beyond the coding stuff, which of course has a lot of work for us to do. But I think there are a lot of interesting things in terms of how we really see that step-function change across the work.

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

My take—my prediction, actually—is that in a world where robotics, drones, and AI are everywhere, in 10 years' time we're actually going to have more Dashers doing deliveries, not less. Simply because the pace at which DoorDash is growing and the scale at which we're operating are pretty insane.

I don't know if people know this, but we have over 9 million Dashers doing deliveries, and the business is growing 25% year over year. Fast-forward 10 years: if we want to 5x or 10x from here, where is the supply going to come from? Are you going to have half of America doing deliveries for us every month? That's probably not going to be the case.

There has to be other areas of opportunity. We're going to have to bring in new modalities as well as improve efficiencies within our business. I think you're going to see DoorDash robotics, drones, Waymo's, and sidewalk robots. We're going to see a world where we're going to have this multimodal fleet, and we're going to need to get our hands on every single modality we can.

I think you're not only going to see more autonomy and more robotics, but you're going to see even more humans as well. With the introduction of autonomy and robotics, and the efficiency gains you're going to see over time, I also think you're going to see an even stronger surge in demand as autonomous delivery becomes even more affordable.

I look forward to getting these today. [laughter] Amazing. Andy, when you think about what you've learned with the initial foray into agentic commerce, how are people going to buy differently in the future beyond food?

Yeah, I think one of the trends that I found fascinating is that, over the past couple of years, Google search query lengths have gone longer.

And I think, to me, how I've translated that is, okay: people feel more comfortable talking to agents or apps like they would a normal human being. If we fast-forward and look ahead to the future, the easier we can make it for people to interface with apps or 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, groceries, or retail, what have you. I think another thing that's going to be true is that we're all going to need to think about what the agent-first experience looks like. We've been testing some of that with the recent DoorDash CLI that we launched last week.

But I just think there are a lot of interesting emerging use cases that can crop up once you start thinking about this. One concrete example I can talk about is someone who was really excited to use the DoorDash CLI because they were like, “Hey, let me basically streamline my office manager use case for my startup.” When they found out that DoorDash did more than just lunch, they were like, “Oh, actually, wait, DoorDash can order me convenience and groceries.” So they just pointed a camera at their pantry shelf, and whenever the shelf was getting empty, they would fire off the agent to basically restock the shelf. I think those types of use cases that you wouldn't really think of are going to unlock some interesting use cases that would not really be as feasible or possible in today's world. 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 of DoorDash. Thanks, guys.

Yeah, it's a pleasure to be here. Find us on Twitter at no prior pod. [music] 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-briers.com.