All right, Tony, man, you have me a little nervous. I understand you want to ask me a question.
Well, I feel like it's your inaugural conference, and you're doing all the talking. Maybe we just give you a back seat for a bit, but why don't we put you in the hot seat for 30 seconds? Investors, you're judged by investments you make. You're also judged by investments you didn't make. And especially with someone of your caliber, your know-how, your expertise, your ability to command and cultivate an audience, it would have been awesome if you were a DoorDash investor. You had the opportunity, I'm just saying, eight years ago. You want to tell the crowd, maybe we start there, what happened?
Sure. To quote Shakespeare, "You come not to praise me, but to bury me." Thank you. Yes. Look, it's kind of a meme when investors, particularly growth investors, they say, "Hey, our biggest mistakes are not mistakes of commission, but rather mistakes of omission." I think people make fun of the statement, and clients hate it, but it's true. I, in a former life, I think I had the opportunity to invest in DoorDash at $600 million. Roughly, what are you worth today, Tony?
Depending on the day, north of 100 times of that.
Yeah. So 100x in eight years do not grow on trees. So this was obviously a mistake, and a terrible mistake. But I actually think it is kind of instructive to share the thinking, because then we can talk through, and everyone can understand how amazing and special what you have accomplished has been.
So first, it just was not obvious. Delivery and ride sharing were both obviously going to be kind of scale-based network effects businesses, where the bigger the pool of drivers you had, the bigger the pool of riders, the more you utilize the drivers or the deliverers. You were coming in kind of as a distant second, and you were focusing on suburbs, which at the time, I, along with a lot of other people, thought was a mistake. It turned out to be really, really smart.
So maybe talk about, Tony Xu, why you posterized me. How and why?
I think people love sound bites to these kinds of questions, and the short answer is thousands of tiny decisions that ultimately laddered up into building a better product experience. But let us take some of the things you talked about, right? The business that we are in of local delivery is really what I call a minimum viable scale business, where you need some number of orders, whether it is at the store level, the neighborhood level, the street level, the city level, to make the economics work, and you are investing until that point, and then you start capturing cash flow per order.
That is actually why you start with restaurants. You start with restaurants because it has got the highest number of stores across any category of local retail, and it has got the highest frequency of activity when it comes to anything being delivered. You talked about the suburbs. What matters is less the population density, and I think a lot of people confuse population density with how you can actually generate route density to actually make an impact on the unit economics, which is probably why you were thinking back then about, "Oh, it is a mistake to go into the suburbs."
But it turned out that it was actually faster to deliver in places like in Austin, where you had easier parking and single family homes, or a greater percentage of single family homes. It is pretty hard in New York City to climb 40 stairs or take the elevator into the right elevator bank and find the customer. So those are, I think, some of the high-level points. But at the end of the day, DoorDash was able to build a product that had more than twice the retention of any of our peers.
And so what we needed at the time when you and I met was we needed capital. Whenever you have a retention and a usage advantage, even though you may still be not the leader, it kind of means you have a bigger motor than the other player. And what you need is you just need fuel to power through, and ultimately, when we did raise capital, we were able to race ahead. And I think we predicted within about a month of when we would take market leadership, actually, two years before it actually happened. And it was just this kind of mathematical equation of knowing exactly how you work the unit economics with the scaling plans, and then obviously, we developed into a multi-category company later.
It's amazing that the cohorts could be that predictive out two years. That's actually pretty amazing. That is a testament to the precision, the execution, the predictability. That's awesome. So Tony, some recent news, acquiring Instacart. And I should say, I was hoping Dan Sundheim was going to interview Tony, and after his opening question, I really wish he'd interviewed Tony. But I think Dan owns as much Instacart as he does, there was a feeling maybe it was not the best idea. But can you talk about the Instacart acquisition? I mean, from my perspective, it really makes sense from first principles, but let's break it down.
Yeah. So back to what is the goal of DoorDash? DoorDash's goal is to grow and to empower every local economy. How do you do that? You do that by helping every local business, whether it's restaurants, a grocer, a non-food retailer, be able to grow both within their channel and the DoorDash channel. And DoorDash plays in both of these worlds today. So restaurants obviously was our first category, and I explained, if you wanted to build the greatest density network in a minimum viable scale business, you got to start with restaurants, and you have to do well in restaurants.
When I think about grocery and the grocery opportunity, I think it's actually still very under-penetrated. And the short reason is because, well, let's start with what does it take to actually do well and build a leading product in grocery? You need wide selection, you need affordable prices, and you need fast delivery.
To do that, you have to invest and build lots of infrastructure. You have to create certainly a modern e-commerce tech stack. You have to create the logistics fulfillment capabilities to get the deliveries perfect, meaning you have to have perfect accuracy, which is hard to do because if you're sending a driver into a store to shop for groceries, guess what? They're competing against other consumers who are just shopping there anyways, and those consumers are going to move apples from aisle six to aisle 10, and so grocers don't know their own inventory. So you got to solve that problem and get that perfect, and then obviously you have to fulfill the orders in an expedited way.
When I look at the landscape and the ecosystem, there are a couple of very large companies, some of the largest companies in the world who have invested tens of billions of dollars into building out a great e-commerce tech stack, great fulfillment to offer fast delivery. But when I look at everybody else, every other grocer, and I look at certainly the independent local grocers, they don't have the resources or the know-how to do that. DoorDash right now is investing in technology. We're investing in building warehouses to store the inventory, so the logistics capability to get perfect accuracy, and obviously we have the network to actually deliver quickly. We have to also invest in scale, and that's where the Instacart acquisition comes into mind.
Because when I think about how do you build this ultimately to enable and unlock the potential for grocery delivery, we have to achieve a lower cost structure and invest it back into the ecosystem so that customers can get the widest selection, the best and most affordable prices, and fast delivery.
Awesome. Can we talk a little bit about agentic.
Sure
Instinct, Muse? I think on the call you said several hundred orders. I forget if it's per day, per week.
Well, total, it's still very early. I think that probably got blown out of context or that piece got isolated into that piece without everything else I said. But right now it's still early.
But how do you think about agentic? You've been on your front foot for a lot of these new technologies, and it feels like a lot of retailers, banks, insurance providers are scared of it. How are you thinking of it?
Yeah. Well, I view them as another channel of acquisition, put simply. But when I think about AI or any other channel, I think about why should a customer actually use that channel? It has to add some value, right? For example, this is why when I look at the DoorDash app, when we built Ask DoorDash, our own AI agent, how does it add value? It adds value by helping you create a grocery basket in under two minutes. It helps you find restaurants that are most similar to your go-to orders that you make for your lunch orders or your weekly Friday night orders, things like that. So agentic commerce is only going to work if, one, it adds unique value for a customer. Two, if it actually completes the end-to-end workflow.
I think that's something that is somewhat misunderstood or not thought through carefully about some of the discussion, because I think it's a cursory discussion around, oh, agents are just going to do everything.
I hope so, if they can add value and actually solve the end-to-end problem. When you take a view at the DoorDash problem, where you have an online beginning but an offline finish, you have to solve that entire flow. You have to solve the exact location you're going to meet the driver. If a SKU is missing from aisle seven and it's going to get replaced by an item from aisle 12, you have to solve a returns problem. If the shoe size doesn't fit exactly perfectly or the T-shirts don't come in exactly the same fit that you want, those end-to-end problems all have to be solved somewhere.
When I think about the end-to-end workflow, the best place that's going to happen is going to be through the DoorDash app. Does that mean that the DoorDash app won't change? Of course it's going to change, and of course, we are already investing in our own agentic capabilities to make it easier to use the product to add the value while giving you the benefits of a perfect end-to-end workflow. It's also going to happen off of the DoorDash app.
No one channel gets to dominate or take over the 20 - 25 eating moments that we have in a week. When I look at the shopping moments on top of that, no one destination is going to take it all. That's never happened. That's never happened in any era, whether it's when e-commerce first came online or mobile commerce.
I mean, we can study many historic examples of this. You're going to have people shop wherever they want. We don't fight that gravity. We welcome that. We think about how are we going to build our destination to be a place that solves the end-to-end workflow the best and give users the most reason to come back to us. If we can do that, then everything will be fine.
Awesome. Autonomy.
Yeah.
How are we thinking about this? Autonomy, robots in all ways, DashMarts, Instamarts, whatever we're going to call them. Autonomy, robots, drones, Cybercabs, like big topic.
Yeah. Big topic, complicated topic. I mean, you're talking about several companies that we're going to have to create just to enable the future of delivery. Autonomy is something that we've been investing in since 2018. What I would say is, the first thing I would say is that it was surprising to me that the hardest part about autonomy is actually not just making sure that something can drive itself or fly itself. It's actually the intersection of that technology problem with the real hard operational limitations, constraints, and challenges that we have in delivery. In robotaxi, a passenger can solve a lot of those problems by getting out of the vehicle, getting inside the vehicle, meeting the vehicle, maybe not in the perfect desired destination or location. That cannot happen with burritos and bananas.
They do not move into and out from vehicles, whether it's by air or by land. We have to solve those problems. The way that we're thinking about it is we're taking a very iterative approach with all of our merchants in actually making sure, can we actually solve real problems? For example, can we actually make these deliveries faster? Can we make these deliveries more accurate? Can we make these deliveries lower cost? Until we achieve those milestones, we won't scale those programs nationwide or globally. So far, so good. We've been at it for about eight years. We have two parts of the program. We have DoorDash Dot, which is the world's only vehicle and first vehicle to drive the road, the sidewalk, and the bike lanes. It's basically a self-driving car squished into one-tenth the size.
There are reasons for that to solve uniquely the delivery problem. Then we have DoorDash Air, which is something that we launched this year, which is our drones program. These programs are just part of what I call one part of fulfillment, right? You still have to get inventory close to where customers live. Those are the warehouses that we're building, and that's what enables you to have perfect inventory accuracy. Then those warehouses are attached to either parking lots, which become depots, or ceilings, which become helipads that we manage that allows autonomy to ultimately flourish. All of this is done in combination and in concert with all of our merchant partners.
Can you give us some scale of the Dot and DoorDash Air programs?
Yeah. Well, Dot we've talked about publicly where we're seeing close to 10% of the volume in the Phoenix market being actually completed by Dot. These are not demos. These are real customers, real orders that are flowing through. We've done drone deliveries as well. A lot of those actually have started with partners in other parts of the world, and there's a permitting process here. It's not exactly at your own pace, if you know what I'm saying. That's kind of what I see. What I see is that ultimately, we'll have our in-house vehicles and aircraft. We'll work with also an ecosystem of partners.
I mean, part of why we built our own program was to really understand the details of, A, what makes autonomous delivery actually work at scale, at the promise of the benefits that customers are expecting to receive from autonomous delivery, and then also to know how to solve the operational challenges and put it into what we call our autonomous delivery platform. That's basically the brains behind the scenes that orchestrates every single order. Does it go to a drone? How do you load it from the restaurant systems, the POS systems inside retail stores, grocery stores, all of that jazz, and makes every order perfect?
Where are we? Let's break some news here, Tony. Can we talk a little bit about Dot?
That's not what I was told I was supposed to do here, but—
But Dot, have you shared anything, how that is progressing against your goals of making the unit economics better, a better experience for the customer, a better experience for the supplier, 10% of orders in Phoenix? That sounds like a lot. Are there any numbers you could share or how that's going or?
I mean, the way I answer this is really just how I think about product development on these types of large investment-type projects, where they're large for a couple reasons. They're large just because the dollar sizes are large, but they're also large because you're making decisions sometimes many years into the future. We have to make leases on real estate, for example, a few years into the future. We have to make commitments to our manufacturing partners on components as well as certain subsystems of both the aircraft as well as the land vehicles several years into the future to get the manufacturing capacity allocated.
And to me, this is probably one of the hardest parts of my job when I think about capital allocation. How much risk do you want to take on? When do you want to take on that risk? A lot of how we think about this is we really have to see the proof in actual customer benefits before we make aggressive scaling decisions. We have the benefit, though, of billions of orders that we can actually test from.
So a lot of what's happening right now with Dot or Air is testing the vehicles in different climates, different type of operating constraints and environments. Obviously, we're also working through a permitting process in each one of these types of geographies. And when we see the customer benefit scale, put a different way, when you see a scale, these programs, you should have confidence that the unit economics are there and that customer benefits are real.
Well, one thing I would just say is I always feel when I get food delivered, there's less french fries than I expect. And hopefully, that problem goes away. And maybe it's just in my mind. But I am very excited for this. Something that I've always thought is interesting, so we're an investor in Zipline. Keller was here yesterday, gave his businesses, he's doing really, really well. I actually don't know what has been announced publicly or not, but anyways, he's killing it. But something that I've always thought is very interesting is that you share a board member with Keller in Alfred Lin from Sequoia, and that Alfred, with all the visibility he has into Zipline's amazing progress, I think has been a net buyer of DoorDash. Yeah, it's just interesting thing to think about.
Yeah, I mean, again, I think sometimes as just people, we have very natural instincts to think zero or one. When I think about this, just as we talked about agentic, we're going to have many sources of traffic that come through the DoorDash ecosystem. We're also going to have many sources of partners when it comes to the ecosystem in autonomy. Some of that, we're just talking about one part of the stack here in autonomy, which is the vehicles. We didn't talk about partners in real estate, partners in financial services, and other parts to actually enable this entire industry, as I think about this, five, 10 years out.
Well, personally, I would say I think it's very clear that both companies can win. Do Amazon and Uber, it seems like Uber's doing more and more partnerships. How do you think about that?
There's always a lot of activity out there when you're talking about something as big as the local commerce opportunity. I guess the way I think about this is mostly what are the things in our control? There have been partnerships across the board in our ecosystem in the U.S., globally, ever since I've been running the business from my apartment 13+ years ago to where we are today. I guess when I look at our performance since then, it seems like DoorDash is doing okay. It seems like people want DoorDash more than they want other services.
When you look at the growth rates of our business, whether it's in restaurants or in other categories, when you look at the retention and the usage, when you look at how they've continued to maintain or accelerate after the pandemic, I think that durability of growth doesn't just speak to just how early we are in the opportunity to get everything inside your city delivered, but it also speaks to how the DoorDash team has continued to execute on all the things you talked about at the beginning of this talk, getting the details right, making sure we have the best experience, making sure that we deserve to earn your business.
The way that I've always thought about the DoorDash business is, as long as we can continue to give customers more value, more selection, greater affordability, better quality of service, the rest tends to take care of itself.
Awesome. Can you talk about AI internally at DoorDash? How are you using it? How's it changing the business? What are you using? What's your stack? How's it changing? How spend is trending?
Well, I mean, it's a very dynamic space. I think we were single-digit customer number for OpenAI, like maybe five-plus years ago. I can't believe that it's only five-plus years ago and we're on whatever version we are on today. Put a different way, I would say that five years ago, I kind of made two pushes to the company. One is, there's certainly, five years ago, I thought about this competition for atoms and this competition for bits. DoorDash plays in the first of those battles, where we want to make sure that we can do the best job in building the highest quality and the strongest local delivery network to enable local commerce in any part of the world. That's the game we're playing. That's where all of our mindshare goes.
But on the parts in terms of reinventing every part of that local commerce stack. That's where all the AI has occurred. How do we use AI to achieve greater affordability for customers, greater profits for merchants, greater earnings opportunities for Dashers? How do we do a better job in customer service with AI? How do we then take AI across every corporate function and actually build it as if we were starting the company anew?
I mean, in many ways, I think for every company in the world, AI, and this is why I'm so envious of startups today, is you get to literally build every function from scratch again. That is such a privilege. It's certainly easier to do it when you're starting from scratch, but even when you get my privilege of doing it when we're a bit bigger than that, it's still awesome. You get to recreate the company. It's a refounding moment.
That's awesome. Any stats you care to share or details about the stack you run? Is it mostly closed frontier, American frontier models? Have you fine-tuned open source? Do you use open source? Stats on spend, things like that.
All of the above. Short answer is all of the above. I mean, we want to make sure in a very safe and responsible way that customers get the best quality intelligence at the best cost structure. That's true for anything that we're doing, whether it's offering the best personalized experiences, making sure that a customer's wallets and loyalty benefits and DashPass benefits are always loaded for them, even if they didn't subscribe, in a way that actually benefits them. Making sure that Dashers have the greatest opportunities to know exactly where every single entry into this stadium is, or wherever they're delivering. Making sure that merchants have their online ordering and their mobile ordering perfectly fine-tuned so they can generate the greatest conversion on their own channels. That's what we're focused on.
Any stats you care to share about how your spend has trended and how you think it might trend over the next year?
I mean, our spend this year went up 20x, and then it kind of plateaued a bit as, I think a few things have happened. I think the frontier models have made their products more efficient. There's been the use and incorporation of open source, open-weight models, excuse, in which we fine-tuned and trained ourselves. That combination has put us in a spot where it's going really, really well. I mean, we're always at DoorDash trying to find the greatest intelligence per token or per dollar spent. As long as that equation works out, we will continue to invest.
After this kind of plateau through efficiencies, open source, we see this at a lot of companies who are early adopters, that you can We had Fireworks in here yesterday, and you can fine-tune, you can take your data, fine-tune an open source model, and put it behind a router, and you generally get frontier outcomes at 30%, 40% of the cost. A lot of early adopters have done this, but what we see there is they do that and then spend starts to go up again.
Well—
You think after this plateau, DoorDash AI spend will grow in the future, stay flat, decline?
I think it'll grow because AI is going to enable all of us to build a personal assistant for your city, for every customer. When I think about that, the intelligence required for the tasks that must be completed to truly deserve that spot in your life is only going to go up. So there's always going to be these kind of curves, right, that go up as the intelligence required goes up, as the models get better. Then there might be a period in which there's an efficiency discovery process, and then you do it again. You kind of keep climbing because we always want more as people, whether it's consumers, merchants, or Dashers. As long as the intelligence is producing greater customer outcomes, we will continue to invest.
Awesome. What is AI doing to make the lives of ordinary Americans better?
AI is making it possible for the small business owner, like my mom, actually be able to compete against much larger companies with greater resources and know-how in today's modern world.
What kind of business does she own?
She ran a restaurant for over 30 years where I washed dishes growing up in the middle of the country. When I think about people like my mom, I mean, it is really the reason why DoorDash exists. This is not a job. This is a passion project that describes her entire livelihood and personality and ethos. There is no such thing as a weekday versus a weekend, a holiday versus the normal day. Every day is a day living her passion project. It is a real privilege that I get working with over 1 million of these merchants, who every single day get to express their passion projects to the world, and our job is to carry it forward and maximize that exposure.
What is DoorDash doing to make sure that, whether through AI or your driver relations, you are having a positive impact on America?
Well, DoorDash wants to grow GDP locally. That is the best way we believe that GDP ought to be grown. When I think about what AI can do, that means greater affordability for consumers, greater earnings for Dashers, and greater same store sales and profits for merchants. If we can do that, we can continue to grow the cities and the neighborhoods that we all enjoy to live in.
Yeah. After years of big box retailers taking share, maybe AI enables small business to start taking share. That is a beautiful thing for America. In three years, Tony, what is something AI is going to be able to do that is awesome for ordinary Americans, your prediction, that it cannot do today?
I think AI in the future will become our own personal assistants, whether that means it is your coach for nutrition or health, whether it means it is the person that gives you financial advice or advice on parenting, or helps you shop more locally and discover more of the underdog brands. AI will become our personal assistants, and it will become a greater part of how we will be productive.
Awesome. Great job, Tony.
Thanks, Gavin.
Good stuff, man. I hope you're the only person who posterizes me up here.
It's your conference.