Tony Xu on DoorDash DOT robot, 35-country expansion, and the physical delivery wars

Welcome to the show. Show. Take a seat. How are you doing? Hey guys. Great to have you guys. I wish we were able to capture the conversation we had. Yes. Off. I was locked in. I was writing the newsletter. John was writing. Catch me. What were you guys talking about? But uh yeah, quick quick intro.

Patrick, why don't you jump off? You shouldn't need an introduction at this point, but introduce yourself and then we'll get into uh the news with Jordan. Sure. I'm I'm Patrick Oanosy. I run a media business called Colossus and an investment business called Positive Sum.

Uh the the shared DNA between the two of them is looking for the very best stories to interview the people, profile the firms, invest in the businesses. Uh so we try to do that all year, every year. Amazing. Tony, welcome to the show. Thanks. It's good to be here.

Also don't need an introduction, but uh I say let's jump right into the news from this week. Yeah, let's do it. Uh what's the latest? Well, we had a lot of launches uh coming. That was my criticism. I was like, you could have spaced these out a little bit because you did six major announcements.

We'll definitely think about it. I mean, I mean, as you, you know, mentioned correctly, a lot of this was like many years in the making, but you know, we're always shipping tons of things at Door Dash and um, you know, some things kind of just line up at the right moment.

So yesterday at our Dash Forward event, which is kind of like our annual event, where we give you a behind-the-scenes look at some of the things we're building, uh, we launched several big projects.

Um, one of them is called Going Out, where now you can use the Door Dash app to eat inside restaurants, which is the opposite direction that I think most people tend to know about Door Dash and their relationship. Uh, we launched Dashmart fulfillment services.

So you can think of this as giving every retailer the ability to, you know, do one-hour delivery. um even if they don't have stores. Um uh third, we launched Door Dash Dot, which is the first autonomous vehicle that can travel all surfaces, so the road, bike lanes, sidewalks. Um it's about a tenth the size of a car.

It can go up to 20 miles an hour. Um and fast. It is pretty fast. Um and so it can handle a large percentage of deliveries, especially those under five miles. Um and you said uh is it 40 to 60% of deliveries depending on the the the area?

Yeah, depending on the market and depending on you know uh the regulations um DOT can address up to you know half of deliveries today and how long you've been working on it.

DOT's been something we've been working on for about seven years and so if you think about what makes DOT come alive or why it's so hard um it's it's actually because you have to build several things.

You know I think most people think of the vehicle and they and they say okay that's part of it but then yes that is part of it you have to build the hardware you have to write the software which you know includes the L4 autonomy stack that we've actually created then you have to integrate the hardware into the software um then you have to you know integrate all of what you've created into our logistics system so we also announced our autonomous delivery platform which means that you know regardless of what type of modality for delivery whether it's drones or uh sidewalk robots or DOT um or uh car vehicles um we can integrate all of those um and our systems will actually figure out uh which fulfillment method would be the cheapest the highest quality tie into all the merchant systems.

So one of the merchant systems we actually announced was smart scales which is another hardware device that we launched yesterday where merchants can get perfectly accurate orders by weighing the items before they actually go out. This could be true for retail. It could be true for restaurants.

For restaurants, we're already seeing 30% improvements in accuracy. And so all of this is tied to point of sale systems and everything that actually happens inside the stores. Does Wall Street understand this? Like I I remember I interviewed Tony for the first time 5 years ago, something like that. Yeah.

And when I asked around amongst friends that covered the business, the common thing that you'd hear was that Tony's ability to get in and the business's ability to get into like the fine grain tiny details that like all all tied together was unlike anybody else.

But I don't do you think Wall Street understands like the the compound impact of all of this stuff chained together and what it will do for the business? Well, I think like most people uh we always underestimate compound interest.

And you know, I I I think there's a couple things sometimes that are less well understood about Door Dash and and you know, some some of it makes sense just because of our evolution, which has, you know, changed quite dramatically over the last 5 years.

So, for the most part, I think most people still think of Door Dash as as bringing you lunch and dinner and doing that exclusively in the United States.

And you know the last 5 years we've morphed from this single restaurant delivery use case into growing in the US into you know being on track to being actually the largest outside of restaurants too.

So we're on track to being the largest third party marketplace delivering all non-rest goods um in addition to restaurant goods. Um we are live now in 35 countries so 34 outside of the US gaining share in every geography. We have um a B2B business that serves hundreds of thousands of restaurants and retailers.

So, if you've ever used the McDonald's app, if you ever use the Starbucks app, there is Door Dash software in there that helps them with mobile ordering, that helps them with delivery. Um and we have an ads business that was the fastest in history to achieve a billion dollars in annualized revenue.

And so now, you know, there's like get that gong. We got to get the gong. All right. There you go. But like so we have those five businesses but we now are launching you know new businesses. That wasn't enough you know. So I mean think about it.

I mean yesterday we launched our autonomy efforts, our instore efforts which is you know helping um consumers connect inside restaurants and inside you know in the future stores. Um and then our Dashmore fulfillment services which really enables you know anybody to have Amazon Prime like capabilities. now. Yeah.

Um even if they haven't invested in the infrastructure themselves, we're building it for them end to end. The warehouse, the inventory, the logistics. Makes sense. Uh when you started the company, there was a boom in mobile cloud.

It feels like that like that was like a unique technology that was unlocking uh the business. Then we've talked to a lot of entrepreneurs who have been at it for a decade plus. And at a certain point, it just, you know, the growth sets in, it gets tiring.

Uh, but I've talked to a lot of founders that have been completely reinvigorated by the AI wave. Does your energy level with your like experience of the company track with that kind of like plateauing and then you're reexited or have there been other ups and downs?

Well, I think I think one thing to kind of piece out of here is there's two kind of wars happening right now and this is one of the things we were covering off the air.

You have autonomy which is underhyped but is plays into this and then you have AI and agent commerce which is getting a lot of attention and so yeah I think Door Dash can play into a couple of those things but obviously the I'm also so curious about the intensity of those wars like if there's the physical wars that you're in against some of the great companies in the world Amazon Mtoan in China others how intense does that war feel today relative to three years ago or something like that just like give us a Yeah I mean I I think there's two like two questions, right?

One is you're I mean you're definitely right, John. I mean like there is what I call like founding moments in a company's life, right? And it it doesn't have to take an AI movement or a new tech shift to get there.

Um even in Door Dash's own maturation, growing from one product into multi- businesses, 35 geographies, you know, I think that alone, you know, gives you reasons to refound the company. um because you need, you know, new leaders, new talent, new resourcing, um uh new efforts.

And so I think you're always building the core and building the new when you're trying to build something over many generations. On the question around these two wars, I I think that's totally right. I mean, you guys were just talking about some of the announcements today.

Um and there's one where everyone's trying to be your digital assistant. There's another where everyone is trying to be, you know, the network physically to make things happen in the real world because you need both of those things.

If your if your digital assistant can't actually get done any of the actions you need on a daily basis, it's not going to be that useful, right? And so I would say that war has always been very intense, Patrick. It's just less noticeable because it's usually behind the scenes. It's physical.

You know, when we're talking about things like warehousing and infrastructure, those are not really consumerf facing assets. Um or even when you talk about things like dot where you have to build multiple technologies, component parts, those are not as um you know quickly seen um or or quickly even deployed, right?

Because you need things like permits and you got to go through regulatory agencies before you can just launch versus shipping something digitally. DOT's go to market. You're live in Phoenix right now and then what what is the next 12 months, 24 months, etc.?

So DOT's been live doing real deliveries, not demos or anything. Real deliveries, uh, hundreds of thousands of miles already in the Phoenix area. We'll cover about one and a half million people, probably something like that.

Um, the the tricky part, you know, and this goes against to some of the earlier question about how intense these things are is that there there is a bit of a cadence that's hard to manage and forecast, right? So on the one hand we have to make commitments to manufacturing partners um to build for many years ahead.

On the other hand it's unknown the order in which we're going to get permits and you know the allowances to launch in different cities. But I can tell you the wait list from merchants is very high.

I think everyone sees the promise and potential of autonomous delivery especially when it's done right especially when it's done seamlessly integrated into all of their systems which dot you know from day one is built um purposely for um but you know that's that that's that challenging management problem that when it's eight years in the making like this so now you get to enjoy the spoils of all the work you did I don't it doesn't feel like spoils Patrick um so if you think about what it takes to win in the physical wars I'm curious like who the other most formidable competitors are I'm sure Amazon's one of them as an example, but in in the back in the culture, how do you make sure the business can keep winning the war and keep winning the marginal battle?

Like what is it in the Door Dash culture that allows you to do this so consistently? Yeah, it's a good question. I mean, you know, back to this concept of founding moments and refounding moments in a company's life, you always have to do two things.

And I think you know and and they both are actually um equally hard and require completely different management perspectives and and and cultural points of view.

One is this idea of you got to build the core and that's where you know every minute shaves matters in a system like Door Dash is for instance we launched uh our own internal mapping system yesterday in a completely redesigned Dash app that took that took over five years to build where you are fine-tuning parameters like gate codes parking spaces apartment you know points of egress and ingress and and um and all of those things matter and they add up to making the core service just a little a little bit faster, a little bit more accurate and help, you know, it helps stashers earn more.

In addition, in parallel, you also have to have a separate management system and playbook to build the new.

So, you know, on autonomy, which we started seven plus years ago, um or uh our our network of warehouses, Dashmarts, now you know, rolling them out as a service, which started 5 years ago, you do have to think about the future. And there it's it's not about optimization.

It's about how are you going to create the new and how are you not going to be delusional along the way because sometimes you know you can have these grand urs and they turn out to produce nothing of of actual customer value and so on on the new it to us it's about you know which problem are we actually solving like with dot we actually didn't start with the premise that we need to go build something ourselves we started with the premise that we believe here are the following requirements to solve autonomous delivery which are different from robo taxi And we went out to the market, talked to and actually partnersh have to solve from first principles all of these things.

And then along the way, it's how do you derisk that?

because in something like DOT there's a lot of capital uh usually at stake and in and and making intelligent decisions to find product market fit that we actually have confidence to hit the next milestone before we deploy a lot of capital is is autonomous delivery more of a hardware problem or a software problem like it's a really uh hard and good question right I mean I I think in um and and I can argue both sides of this I think in Silicon Valley we tend to think of most things like software problems mostly because we get to control them and we get to deploy super fast and you know billions of people use our products um uh in a short period of time.

That's not as that's not as true in in in hardware. Um and as as as a result um I think we tend to underestimate the challenges of you know manufacturing and how do you go from a demo video to real deliveries to you know every 10x in scale. Um so that's the challenge on the hardware side.

On the software side I think what hardware companies you know maybe traditional automakers and other companies tend to underestimate is you actually do need to invest um quite considerably in software whether it's building or fine-tuning.

um you know your LLM for the physical world collecting lots of data you know a mileage in the case of DOT um so that you can actually integrate the two so the third part of the integration is non-trivial and then there's the fourth part which we didn't even get to which is around operations right so how does you know how do we actually solve the last 10 feet you know in the case of robo taxi you can get in and out of your whimo yourself in the case of dot someone has to load and unload the vehicle That's why we built DOT so that it doesn't just go on the road.

It can also go up to a doorway that stood out to me is if if you were a hardware provider that just wanted to make autonomous delivery hardware and then you guys are sitting there realizing like Google Maps isn't actually even good enough to do our existing delivery process. We need to build that too.

And that's kind of the Yeah, there's a mapping system. There's a there's a merchant systems integration system around kitchen display systems, point of sale systems, um you know, weighing systems to you know for order accuracy.

Then there's a dispatch system where you know sometimes um you know dot you know could get the job done. Sometimes maybe it's a sidewalk robot. Other times it might be a dasher a human dasher. Other times it might be a drone and you know all of those things.

So the operations and then there's the maintenance and the repair and the rescue and the tea ops and so to do something like dot why it took seven years to get to a v1 that's why I wouldn't call it you know reaping in the rewards right now it's still you know somewhat pain and suffering um it's uh you know the team deserves a ton of credit of even getting to this model can you imagine trying to compete with Tony in the physical wars I'm curious like if you were a perfectly evil competitor wellunded how you would compete with yourself?

Like, how would you even try to attack Door Dash's position in the physical wars? Well, I I you know, maybe I probably answer the question how um I answer most of these questions, which is to me like what motivates um what I do is how do we just keep getting better?

Like it's not a race against other people, you know, I like for example, take the case of um something as you can't see them anymore. No, take the case of something as boring sounding as like, you know, accurate deliveries, right? Um, we do 8 million plus, you know, deliveries every single day.

We are definitely not going to be perfect today or even yesterday or tomorrow. Um, that means we always have room to improve and it means that the clock resets tomorrow and gives us a chance to get to perfection.

And I think that's the kind of DNA it requires to have a chance at I think winning in the physical world which is very different right in the in the digital world hallucinations are okay you know some um not all the time but like they you know users users will give you users will give you yeah you can have a a generative generative AI product today that is produces something terrible 15% of the time and that can still have incredibly incredible product it'll be totally fine and and I think in the case of, you know, the physical world, you kind of have to be right.

And um and and and again, that's why I think there's going to be in the future a very nice marriage and blending of of these two worlds coming together to make the most productive use for all customers. How do you think about the uh market structure in the early days? And has it played out the way you thought it would?

It feels like some markets just become duopies, igopolies, some are more monopolistic. Like what were you thinking? Were you even thinking about that in the early days in the early pitch decks like winning market by market?

Yeah, I was I mean I had a point of view on what the structures would be mostly because if you look at businesses like Door Dash, they're what I call minim minimum efficient scale businesses.

What I mean by that is you have to achieve a certain amount of scale in in order to even have a chance at you know lasting as a company in terms of economic viability. So there's a natural moat around economies of scale.

a natural there's a natural I mean I could I could literally derive this for you mathematically like like why for certain markets you need certain volumes and then what tends to happen is I think as companies like Door Dash continue to grow in the core and we're also innovating in the new um I think that just adds to you know the amount of advantage that occurs you know because we're constantly making the economies of scale produce more value for customers and at the same time we're reinvesting all of those advantages or those gains back into making the next set of products.

Are there things that you now think are possible, you know, five, seven years from now that you wouldn't have said are possible 5 years ago? Like how is your view of the scope of what Door Dash might do for people evolved in the last couple years? Yeah.

So I I think um a couple things, one has to do with Door Dash more internally driven and the other has to do with what you were talking about um you know, John earlier around um you know, AI.

So on the on the first question um around you know as Door Dash has become cash generative which has been many years now in in a row um we have more ability to actually invest a a and and as we as we you know gain more volumes um invest in more things. So that that that's that's uh one piece of it.

Um the other piece of it though is you know I think uh given how fast a lot of the advancements in things like reinforcement learning have been I mean even in the last 9 months let alone the last four or five years um it's almost as if um a lot of these problems that we once thought were really difficult um can be solved using different versions of transformers and neural networks and I don't think technically speaking, you know, that was always the case or the even the point of view 3 years ago.

Um, and so I do think that some of the technical advances have given people new ways to solve the same technical problems and also frankly new ways on how to run companies. You know, we didn't really get into that.

I don't know how much time we have, but I mean like you know there even how you run companies now can be very different. The obvious example is coding.

engineers are certainly much more productive today and you know most new code I I would I would imagine certainly for our case but you know many other companies are most new code is written completely by um agents today and I think there's going to be versions of this in every function it it's not going to be perfect and there's there's going to be bumps and and some things are not going to work um but that's just like a new way of Is there another surprising example of that that you like that's actually happening at Door Dash?

Yeah, because you'll see CEOs today, they'll they'll they'll leak an internal message that's I want you all using chat GPT a lot and that that qualifies as like AI native.

Clearly, everybody that gets value from different LLMs knows like it's might be helpful, might be able to rephrase an email better or generate a document more quickly, but that doesn't that it feels like that's still below the bar of being a Yeah, it's definitely not the bar. I mean, I don't even know what bar that is.

That's quite low of a bar, but like but but like I would what I would say is this that there's a couple things that are hard and then I'll try to answer Patrick's question on specific examples, but like you know one of them is this um this comparison of how much easier it is now to be productive versus like what the demand for that activity was.

you you know one of the reasons why for instance it's hard to why I think you know it it's been hard to get as much productivity gains even in coding is because not every hour of the day spent by a software engineer is spent coding um that's only a a small number of hours actually and so um you know the weighted average gain you know for that function versus the demand for the amount of code or new features being developed you know actually doesn't it kind of nets even and so not every company's necessarily seen you know massive gains unless it's a very small company and and you know everything is new and therefore you know the majority of things produced are uh being written by agents but but I do think that um you know this is still like the new way of how you run companies and I think that we're obviously all trying to figure out what are some of the discoveries again back to you know Patrick's question um a lot of where we've seen gains in using um some of these techniques techniques uh are really in um kind of operational processes.

Um so for example, one of the things we have to do is we have to label hundreds of millions of items um that are in the city. No one has that data. That data doesn't exist on any search engine, any you know chat assistant. These are just items in stores on menus. Well, what's the world? We're in LA right now.

what's the be uh how easy is it to find you know a parking space right outside this office or where is the last you know Apple Airpod in you know this you know particular skew or um you know how many pink lady apples are there versus honey crisp apples right and these are you know pieces of information that again sound boring or benal but they are the things we obsess about at Door Dash because we have to right or but if you're a digital assistant those are not things necessarily you care about and and and that's not information you can just you know crawl the internet internet and train on.

And so you know a lot of that labeling for instance um would have taken many human hours um to make sense of in a way now that actually uh can be done with LLMs.

And so that's an example of of of something happening where there's true productivity gains like that that would I mean frankly be a thousandx of what we could have done with just a human operation. What is your vision for Door Dash's role in even the average American's life in a decade from today?

Because clearly it's evolving and No, I love that question. I mean like I I think most people, you know, I I don't think know enough about like why we started the company. We started the company not because we were hungry one day and wanted to get something delivered.

We started the company I mean honestly because we wanted to help business owners like my mom and you know when you think about the local businesses small, medium and large including franchises of big companies like McDonald's or other places um they're very much like my mom for whom you know they have 17 days of cash on hand running a business.

This is their livelihood. if you took it away from them. It's not about like losing a job as a in in a corporate, you know, setting. This is about losing your identity in some ways.

We want you ordered you ordered a coffee this morning for a local cafe and they're not going to be investing in autonomous delivery, but meanwhile big retailers will, right? Yeah. And so we want to live in a world where all of these businesses, physical businesses, small, medium, and large, can actually make it.

And I and we don't want to live in a world where only one or two of, you know, retailers make it.

We want to live in a world where the maximal number makes it and if that happens then the cities are going to grow organically their GDP the job indexes will naturally grow and life will be great right and so for the average person you know we want to make their life a little easier if they're a consumer if they're a business owner we actually want to make them survive and be very very successful and then if they're a dasher or someone looking for extra work we want to actually you know help them out you know the average dasher only works three to four hours a week that's because 80 88 8% of them already have full-time jobs and and so if we can be a stepping stone in their way to achieving some goal, that's what we want to do.

We so we actually want to build technology to solve real problems for real people. Fantastic. Well, thank you so much for stopping by. Uh this was this was really enjoyable. Thank you. That was fun. I I enjoyed having Patrick's voice coming through coming through my headset.

It feels like I'm in an episode of Invest Like the Best. This is great. We know you have to get out of here. Incredible progress. I I I I have one question for Patrick. We can keep him around. Okay. Tony, good to have you. Thanks, guys. [Applause] Chat was asking why we have Patrick Ashanosy here.

If it's social connection, why not? Uh, but I did want to get your take on this Jeremy GeFon post. He said inorganic growth, acquisitions of competitors and suppliers, buybacks, debt are all going to become much more native to the