Today, I’m talking with Uber CEO Dara Khosrowshahi. We recorded this conversation the day Uber announced a big set of product updates, including new options for shared rides and some features to make commuting easier and more predictable. Dara was in New York for all that, so he came to our studio, and we did this one together, which always makes for a great episode.
As it happens, the traffic in New York that day was truly terrible — so we started by talking about how often Dara actually takes an Uber, what that’s like for him, and what it’s like when he goes and serves as an Uber driver, something he does regularly. There’s a lot of security around an Uber Eats delivery when Dara’s behind the wheel, it turns out.
Listen to Decoder, a show hosted by The Verge’s Nilay Patel about big ideas — and other problems. Subscribe here!
Uber might be the single best example of the major service apps that boomed into existence in the early part of the smartphone era. In the simplest terms, I think of it as the button that can make a Toyota Camry appear nearly anywhere in the world. But underneath that simple idea is a complicated dance with tons of dependencies. Since he took over for founder Travis Kalanick almost a decade ago, it’s been Dara’s job to really operationalize and smooth the company out into a stable, profitable organization.
That’s Decoder bait through and through, so we spent a lot of time deep in how Dara thinks about Uber’s various businesses, how they’re split between product functions and the regional teams in all the countries that Uber operates in, and how Uber plans to grow as things like Waymo and other autonomous vehicle companies enter the picture.
Like so many mature service companies, a big part of the answer is to try and become a consistent part in people’s lives that drives recurring revenue – that’s things like Uber One subscriptions, but also bulk trip discounts for regular commuters. And Uber’s biggest news last week was something it called Route Share: predetermined spots where an Uber is guaranteed to come every 20 minutes and pick you up.
Yes, I asked Dara if Uber had just reinvented the bus. His answer is that Uber should be a complement to great public transit, and we talked a little about congestion pricing in New York and how that’s playing out in the data that Uber can see.
If you’ve been listening to Decoder recently, you know that I’m really curious about how service apps like Uber will handle things like AI agents, which promise to let you book a car simply by asking an assistant like Alexa or Siri. What’s in it for Uber to have its service commoditized away behind someone else’s interface, especially as it’s trying to grow new lines of business?
Dara had a lot of thoughts here, and it’s clear that the business side of AI agents has just as long a way to go as the actual tech. There’s a lot in this one, and Dara didn’t hold back. I think you’re going to like it.
Okay: Uber CEO Dara Khosrowshahi. Here we go.
This interview has been lightly edited for length and clarity.
Dara Khosrowshahi, you’re the CEO of Uber. Welcome to Decoder.
Thank you for having me.
Thank you for being here in person. I love doing these conversations in person in our studio here in New York. It’s a different energy, so thank you for coming in.
In-person is the new thing.
I have to ask you, did you take an Uber in New York City to our office?
I did not take an Uber. The reality in my life is that I need security and all that stuff, so sometimes I take Uber and then the security vehicle follows me. Today we had a lot of traffic, so I couldn’t do it.
Yeah, there was a lot of traffic today.
I always wonder how often you get to dogfood your own products.
You’ve got to dog food your own products. Absolutely.
Tim Cook said in an interview recently that he uses every single Apple product every single day, which mathematically seems very challenging. Do you take them all? Do you take UberX and Uber Black?
Totally. You can’t do it every single day, but I use Uber. I use Uber Eats. Actually, one of the really important moves I made was starting to deliver. Most Uber employees use Uber as consumers, but not as many use Uber as earners, as drivers, or as couriers. Early on when I joined, we were building more for the rider or the eater than the earner.
So when I was in San Francisco during COVID, I was going crazy at home. So I got my e-bike and I started delivering food. Then, I got a Tesla and I started driving folks around. I really do think that it’s important to dogfood. You can’t do it every single day, because you have a day job. But for one, you learn about your product. Just as importantly, you’re setting an example for your employees.
When you were driving people around for Uber, was security in the car with you? How does this work?
They were tracking me, and then they were following me to make sure everything was okay.
This is the most intense Uber ride of all time.
It was really cool. By the way, it’s a lot harder than it looks. When I first started driving, I was so nervous. I didn’t want to screw up. I didn’t want to take the wrong route. It’s actually a lot more challenging than you think it is.
There’s a lot to talk about today. You have some news from your Go-Get announcements. You have ways for people to use Uber more consistently, which I think is very interesting. There’s a partnership with VW to launch autonomous rides in Los Angeles next year. That’s on top of all your other autonomous partnerships. I’m obviously very interested in that.
I want to start with Uber just conceptually right now. If I think of Uber in the most reductive way possible, it’s an app where I open it, I push a button, and a Toyota Highlander will show up anywhere I am in the world.
Pretty much, or another car. For you, a Toyota Highlander.
It’s almost always a Toyota Highlander.
That’s awesome. I love it.
There’s a whole lot to that. But in the most reductive sense, I think that’s how people perceive Uber. It’s a button that makes a car show up.
I think so, yes.
Is that the foundation of everything else, or are there more foundations?
Ultimately, we want to be your everyday app, kind of like your iOS for everyday living. We started with rideshare, and that was the core of our business, but then we’ve expanded into other categories. Obviously there’s Uber Eats and grocery. And to some extent, I think we’re building this real-time logistics network.
It started with moving people around. Now we’re moving things and food around, and we’re expanding into many, many more categories. But to the consumer, we want to be known as the app that makes your day a little bit easier, that helps you go where you want or get anything that you want.
The news includes a bunch of things like Route Share. There’s more ways to use the subscription more consistently, more ways to plan commutes to work. So that core function — that there’s a supply of people with cars in the world and Uber can aggregate the demand and push the button and the cars will come find you.
That’s the thing that appears to be changing in your news. It’s not “push the button and the car comes to you.” Instead of commuting, Uber will just manage the logistics of that for you and you will subscribe to it, and that will be a recurring fee. Is that the change here?
To some extent, what we’re doing is saving you time with someone else doing the driving instead so you can do whatever you want. Now, there’s a trade-off as it relates to reliability or time and price. The faster you want your ride or the more reliable you want that ride to be, the more it’ll cost us.
If you say, “You know what? I may not need my ride in the next four minutes. I can wait 15 minutes. Or instead of having the car come all the way to me, I can walk two blocks if that makes the network more affordable or I can save money,” then that’s a trade-off you can make. Human beings are always trading off. Essentially you’re paying for your time. With our network, there’s always a trade-off between reliability / time and price.
So, what we’re trying to do is find products that allow people to self-identify where they want that trade-off. For example, you have a really important trip and are going to take a flight. You need to absolutely know that the Uber is going to be there. You can pick a reserve. You pay a premium for that reserve. The driver shows up early. The reliability is like over 99 percent. That’s one version.
The other version is for the commuter. This is someone who is going to work every single day using Uber. The price of your commute every day can get up there. So, what we wanted to do is allow somebody to trade off a little bit of price for reliability. For example, we have one product called Routematch, which has set routes coming every 20 minutes, so there’s predictability there. You will share your ride with up to two other people, but you are trading off a little bit. You can’t quite get your ride on demand because there’s a schedule there, and you may share with somebody. You’re trading off a little bit of that personal reliability or time in order to get about 50 percent off.
So, we’re constantly making trade-offs as it relates to our services. Originally, when we were thinking about these trade-offs, we thought it was about the demo. There are some people who are less time sensitive and more price sensitive or the other way around. It turns out that that is true, but there are also occasions. An occasion might be travel, where you’re going to pay more to be more reliable. There may be another occasion where if you know the 20-minute increments, you can adjust your life to get that 50 percent off on your commute. So it’s been a really interesting evolution and how we think about the trade-offs and the efficiencies of the network.
Is convenience on that list of priorities for you? You talked about time and reliability, but it seems like if I open the app, it being inexpensive and right there is as important as all the other stuff.
I think your definition of convenience may be different. Something that’s cheaper may be more convenient for a person. Something that’s more timely can be more convenient for another. Ultimately, I think convenience is an amalgamation of all of that and it depends on what you consider to be convenient.
One of the reasons I asked that question is because in various other parts of the tech industry that we cover, consumers will pick convenience over quality almost every time. The example I use, which is not a one-to-one to Uber, is music. Consumers will pick just the worst quality streaming if it’s available. They’ll watch bootleg YouTube videos of the songs they want to hear. They often will not pay for the high-quality lossless elsewhere. And you can do price discrimination based on that.
I challenge you a little bit on that because in the early days, bootleg music, bootleg movies, etc. were the biggest use case. But actually as streaming made it more convenient, people were willing to pay. So I do think there’s a price / convenience trade-off and it’s personal. I think habits change as well. We are just trying to figure out what’s the most optimal price / convenience trade-off that we can make to aggregate the most demand and the most supply.
One of the reasons I asked about convenience is that you framed Route Share or walking a little bit to get to the spot as reliability. To me it’s actually convenience. I’m going to trade a little bit of convenience and maybe pay a lot less.
You can totally go there.
I read this press release announcing Route Share, and I had this very mid-2010s reaction, which was what if Uber just invented a bus. Did you just invent a bus?
I think to some extent it’s inspired by the bus. If you step back a little bit, a part of us looking to expand and grow is about making Uber more affordable to more people. I think one of the things that makes tech companies different from most companies out there is that our goal is to lower prices. If we lower the price, then we can extend the audience.
There are two ways of lowering price as it relates to Route Share. One is you get more than one person to share a car because cars cost money, drivers’ time costs money, etc., or you reduce the size or price of the vehicle. And we’re doing that actively. For example, with two-wheelers and three-wheelers in a lot of countries. We’ve been going after this shared concept, which is a bus, for many, many years. We started with UberX Share, for example, which is on-demand sharing.
But this concept takes it to the next level. If you schedule and create consistency among routes, then I think we can up the matching quotient, so to speak, and then essentially pass the savings on to the consumer. So, call it a next-gen bus, but the goal is just to reduce prices to the consumer and then help with congestion and the environment. That’s all good as well.
Congestion is interesting. We’re talking in New York City. We’re a few months into the congestion pricing program, which was designed to get people out of their cars and onto the bus, mass transit, even into Ubers and taxis. Have you seen any statistics since congestion pricing has been in effect that says Uber usage has gone up?
So Uber usage has gone up, but we don’t have the counterfactual. Would it have gone up if there were no congestion pricing? But what we’re seeing is very similar to some of the public data out there, which says traffic is flowing faster. I think The New York Times said traffic speeds were up like 6 or 7 percent. Uber has very similar numbers as well.
So, Ubers are getting there faster, but I don’t have a counterfactual. Was it good or bad for our business? We supported congestion pricing. The way I think about it is that it’s actually taking cars or vehicles that have the lowest utility off the road. It’s a little bit like surge pricing if you think about it, where during peak times, we want to get more drivers out there.
At the same time, we want to encourage people who may not need to travel during peak times. We actually want to dampen demand and move it to non-peak times. So, congestion pricing to some extent is a “city surge.” They’re increasing prices for a certain time, and it removes demand for vehicles that have the lowest utilization. It just so happens that Uber’s taxis are very high utility vehicles on the road.
You’re closing in on a decade as the CEO of Uber. The way you talk is very operational. You’re maximizing the utility of these assets on the road, whether your drivers own them (we should talk about the autonomous fleets that seem like they’re going to come into existence). Then, there’s maximizing utility, value, and all this stuff. You took over from a founder who had a much grander vision, which was that no one should own a car and that all transportation should be run by Uber.
Well, that’s still a vision. Absolutely.
The way you’re talking about it, likeRoute Share competing with the bus, is it still the vision to take over from those things? Do you perceive yourself to be in competition with public transport?
No. Actually, I view that we are in competition with personal car ownership. That was Travis [Kalanick]’s vision as well. Public transport is a teammate. Uber consistently does really well in cities where there’s significant public transport because — and it goes to what I was talking about earlier — there are just different use cases. You might take public transport for your downtown commute, but if you’re going to dinner with your friends, then you might take an Uber.
So, Uber and public transport compete for a particular trip. That’s absolutely true. You have to decide, “Am I going to take the bus, am I going to take the subway, or am I going to take an Uber?” As it relates to your life, we’re very much complements.
The big kahuna that we’re going after is car ownership. That ultimately is a vision. It remains a vision. It’s hard to do because the car is a really flexible product with mass manufacturing and it’s really cost-effective. So, it’s going to take a lot of work to get there. We’re less than 3 percent of miles traveled on the road even though we’re a very big company. We’ve got a long way to go, but the vision definitely hasn’t changed.
Are there some markets where you think you’ve made a bigger dent in car ownership than others?
I don’t know if we’ve made a bigger dent, but there are definitely some markets where our penetration of miles traveled is higher. For example, in Latin America, our penetration is very high. Penetration in New York or other big US cities where there is mass transit is higher as well. I haven’t looked at whether we’ve made a dent in ownership.
Another way to think about it is there’s some percentage of people who would’ve gotten a car but Uber exists in their life. Do you track that? Do you measure that?
We don’t track that. But I can tell you — this drives me crazy. My son is over 18. I don’t know about you but did you get a license the minute you could drive?
Oh yeah. At 15-and-a-half, in Wisconsin.
Exactly! It was just such a thing. It was a goal in life. It represented freedom. I’m still trying to get my son to get his driver’s license, but Uber’s freed him up. If you look at the percentage of 16 or 18 year olds who are getting their license, that percentage is coming down significantly. I think it used to be two-thirds, but now it’s probably in the 50 percent range. I could be wrong. So, it is absolutely having an effect on car ownership.
Do you think that’s a leading indicator for Uber success, or is that a lagging indicator of Fortnite success?
[Laughs] Probably the two are mixed. I also think it’s an indication of the urbanization of our populations, but I haven’t actually looked into whether heavy Fortnite players are Uber users as well. I suspect they may be. Certainly Uber Eats users.
I’m going to leave that one alone. There’s another hour of PhD-level sociology in that comment alone.
Correlation, causation, who knows.
Let’s talk about Uber itself. You took over in 2017. You took over a company led by a very strong founder who had a vision of how the company should be run. How is Uber structured at this point? This is a classic Decoder question.
We have a combination matrix and line of business structure, so we have two global leads: one for our mobility business and one for our delivery business. Then, most of the other functions — marketing and PR, government relations, product finance, etc. — are matrixed. I would say that as the company has matured, we have matrixed more, which I think happens to a lot of companies that mature. Their practices become more predictable and they’re trying to bring a higher degree of expertise into those practices. So, it’s a combination. We constantly have this creative battle between the lines of businesses and the matrices. Ultimately, I think it works.
I mean, that’s the tension. I think the listeners know that the secret of the show is I can just guess at 80 percent of the problems if you tell me the structure of the company.
Listen, I think that the conflict is constructive. This is a terrible metaphor, but if you look at purebred dogs, they tend to get sick more, like bulldogs that have been bred year after year after year, versus hybrids that aren’t pure breeds. To some extent, I think that companies that have a structure that’s one note — there’s a line of business (LOB) structure or a matrix-only structure — there’s a lack of conflict that sometimes allows the strength of the weakness of either structure to go overboard. Matrix structures are usually associated with more efficiency, but slower decision making. LOB are sometimes much faster but have difficulty scaling.
I think my directs and some of the teams get frustrated with conflict. Some of these decisions come up to me. But I think the conflict is actually really healthy. The conflict is what allows us to not fall prey because these two teams are constantly keeping each other in check. If you’ve got a matrix function, it may not realize that, let’s say, the marketing in Egypt is not sharp enough, but the Egypt GM is living and dying by that marketing. So the two keep each other in balance. That Egypt GM might not know what an absolutely first-rate marketer is. That conflict keeps those two in check and ultimately gets a better outcome. Conflict can be unpleasant, but I think the fight makes us better.
As a guy who runs a single vertical line of business in a matrix company, I understand what you’re saying.
There’s no perfect way.
Put that into practice. You have a hypothetical there. What’s an example where you have had to mediate that conflict?
I think marketing is one, since we just talked about that. I’ve had to mediate as it relates to, let’s say, the allocation of capital. Every single line of business and GM is going to want some budget for branding because they’re fighting it out on the ground. Then, if you look on a global basis, how do you allocate that marketing spend? The answer to the global solution may be different than the answer to the request of that Egypt GM. Those answers during planning may be different.
Early on as we were building this matrix structure, there was a lot of conflict. Now, the teams have a real understanding. They’re getting into a rhythm, and the conflicts get to me much less often. Same thing with product. Product prioritization is very similar, with a constant push and pull.
As a global company, you want to build globalized products that are as similar to each other as possible. At the same time, there are local competitors that may have some product twists and turns that may be only applicable to that market and may not make sense for you to build on a global basis. That push and pull is a tough one sometimes.
When I say 80 percent of the problems, that’s the one I can usually guess at with 100 percent accuracy. How do you solve that? Do you have local product teams that are allowed to just make Uber whatever they want?
We have the GMs on the ground who really keep the product teams accountable, but our product planning is global by nature and it is not perfect. Sometimes the GMs who yell the loudest get their way. We want to make sure that our product people travel to the market so they feel the pain of the market, so to speak. Honestly, it’s something we have to keep working on. We don’t have it perfect by any means.
There’s a difference there from the founder-led company that is totally functional at the beginning with gobs of VC money coming in. I mean, that’s the early Uber, right? We have a bunch of zero-interest-rate VC money and we’re going to win by just buying all the drivers in a market, setting the prices to zero, and we’ll figure it out in the end. And that was totally functional. You’ve created it in a different structure. I’m just wondering, how did you make that transition?
Let me correct you there. It wasn’t functional. It was GM-led.
Even back in the day?
The GMs were mini CEOs of every single country. Travis would hire a GM and parachute them into a market. You’re right that we did go out and buy supply. As a company, we really are supply-led, where first you build liquid supply and then you invest in demand. In the early days, you essentially had to price under market and lose a bunch of money.
As liquidity increases, the matches increase, and the efficiency of the market increases, you can start pulling more profits from a marketplace. That was essentially the formula. It might’ve seemed crazy at the time, but I think that Travis and the founding team got it right. Creating liquidity in the marketplace is exactly what allows you to get to profitability. Whoever created the liquidity supply and demand fastest was the one who ultimately won.
So in the early days, it was all GMs spending a ton of money on liquidity. When I came in, I started to globalize some of those matrix functions. It took time. It was painful at times, but I think we’re at the right balance now.
What’s the hardest choice you had to make along that way?
The hardest choice was actually on product. Specifically, when I was around for a while, our Rides and Eats businesses were complete verticals. So you had different ops teams, different marketing teams, and different product teams. Eats was competing against pure plays: DoorDash, Just Eat, Deliveroo, etc. Eats was 10 or 15 percent of the business.
So if you decide to matrix that function, for example, is your head of product going to spend time on the 85 percent for Rides or are they going to spend time on the 15 percent for Eats? And no matter how much I said, “Hey, spend time on Eats,” they were getting paid for Ride’s success. So, we went a few years with Rides and Eats essentially being totally separate.
Then, there was a big decision that we made to combine the product teams and the product and tech teams. That’s what makes the company go. There’s a lot more that happens around it, but if you don’t have great product or you’re not strong technically, you’re going to suffer. That decision was probably the hardest decision. It came with some bumps and bruises, but I think it was a really, really good decision. For one, Eats got big enough for everybody in the company to care about it, once it got past that 15 percent.
Second, we were late to the game with Eats. We were years behind the other pure players. So if you ask yourself, “Why is Eats going to win?,” we obviously had great talent, but you have to assume your competitors have great talent as well. The answer was the platform. Uber as a brand was an everyday use case, and you could actually introduce your audience who are using Rides on a daily basis to Eats. We then launched a membership program that worked for both.
The platform and the fact that we have greater scale than anyone else allows us to invest more and then cross-sell our riders into eaters. Now more are eaters into riders. Then, encourage drivers to move people but also move things every once in a while.
I want to ask you about cross-selling on the platform in a very specific way. but you’ve walked right into the other big Decoder question I ask everybody, which is about decisions. You just described a big decision. What’s your framework for making decisions?
One framework that I use is, “is it a one-way door or a two-way door?” That’s classic Amazon. I think Jeff Bezos introduced it. It’s a great framework.
One day we’re going to make a supercut of people saying that on the show.
So many people have been impacted by him, and deservedly so. He’s one of the great CEOs of the world. Just make the decision. Use your instinct. It’s always a two-way door, that’s easy. But that’s an example of a one-way door.
You mean merging the product teams?
Yeah, merging the product teams. You’ve got to go for it. I can’t tell you that there was a perfect construct one way or the other. At some point, you have to make some of these decisions based on instinct. One of the constructs is what do you have to believe so that the outcome of what you’re doing today is going to change? So, if I keep doing the same thing but I do it an inch and a half better, am I going to get to the Promised Land? Is that just going to take way too much time, or is there uncertainty in getting there where I need a shock to the system?
As it relates to Eats, we did unbelievably well in the early days, but there was a point where we were hitting a ceiling. It was then where I decided that to get to that next level, we actually need a shock to the system, and we changed our organizational structure.
In hindsight, I think it was a really good decision, not necessarily because of the product side. We still have dedicated Rides and Eats product leaders. It was actually more on the engineering side, on the code base. It allowed our engineers to generalize a code base to take some of, let’s say, our marketplace talent from Rides, who were second to none, and move them over to Eats.
It’ll take a little while, but I’ll give you a quick example of some of these unanticipated benefits. When you order an Uber ride, we essentially have to scan the market for all the cars out there and where they’re going, and are they available or not available. We have to match you to a car, and we’ve got to price that ride in less than 30 seconds. We want the ETA to be four or five minutes, and so we have very, very little time to make that decision.
So, the algorithms that are running have to work very quickly. They’re making tens of millions of predictions per second. Make that match and go. That takes certain architecture and a bunch of compute. With Eats, we have the average time to make food at like 10 to 12 minutes. So, you have a lot more time to make that match.
But we had to make all the same calculus. Where is the courier? Which courier should I match you with? How do I price you? The time with which we had to make the match on the courier side was much longer. So we would recalc the marketplace in two or three minutes. With Eats, we essentially make a bid to a courier who can accept the delivery or not. Let’s say for a delivery that I’ve got three bids I can make before the food gets cold. My first bid might be $6. If they said no, the second bid may be $7. If they said “No, I really have to get you the food,” the third bid might be $10. Once we switched the teams over, the Rides team said, “Oh no, we can make this recalc in 15 seconds.”
Now, we’ve got the ability to make 10 bids. So instead of going from six to seven to 10, you go from $6 to $6.50, $6.75, $7. So the ultimate cost per transaction is lower just because of compute, speed of algorithms, and decision-making. I didn’t anticipate that when I made the switch, but those kinds of benefits — in terms of back end code, engineering, just scale — were actually the biggest we saw.
I understand why that’s great for me as a customer because cost comes down. I understand why it’s great for Uber because the margin goes up. Is that good for the courier?
Well, the couriers get more business. For example, our ability to make more bids has allowed us to increase the number of batches. So now the courier is carrying two goods and gets tipped twice close to 50 percent of the time. That was a side benefit for the courier as well, which we think ultimately helps earnings.
The way you’re describing Uber is as a platform. A lot of the announcements are about using it more as a platform: “Engage with us more often in your life.”
It’s a kind of cold platform, but it’s an everyday use case.
The reason I ask about this is because maybe 60 percent of the interviews I’ve done on Decoder this year are people telling me they’re going to build AI agents. You’re just going to talk to the computer and the Toyota Camry is going to show up.
AI is a cool thing.
Everyone is telling me this is going to happen in ways big and small, whether we’re building Model Context Protocol applications at the bleeding edge of tech standards or whether we’re just going to click around on your website using a testing tool, which is a real product that exists in this world. And Uber has been clicked around by these products.
Those all disintermediate your platform. You have to do the hard work of creating liquidity of supply and making sure the Toyota Camry appears, and they own the customer relationship. Why would you ever participate in this? I ask them all this question. They’re like, “I don’t know. The market will figure it out.” I’m like, that’s you! You’re the market. But you’re the other side of the market. Would you participate?
It’s a really good question. We talk about it all the time, and time will tell as to what the right decision is. I believe in running open platforms. I don’t believe in companies that try to fight the course that technology is taking. They always get left behind. I think we should go to where consumers want us to go. I do think that’s all nicey-nicey, but at the same time, one of the advantages that we have is our unique inventory.
Essentially, we’ve got 8.5 million drivers and couriers and over 1.2 million merchants out there. It’s very difficult for a company that has an agent to disintermediate us and go direct to the almost 10 million pieces of inventory, if you want to be impersonal. These are people, these are businesses. It’s very difficult for them to draw that inventory. While you may make your inventory available to these agents, you can charge a toll if you have unique inventory. So, if the agents bring us more demand, which means more orders for our restaurant partners or more rides for our drivers, that’s worth something for us.
At the same time, because we’re becoming more and more of an everyday use case — the frequency of our average user is six times a month and it continues to increase — we can create our own local agents. We know it’s your commute time, so why don’t we get you a car? Maybe we can actually make a dispatch to the car before you actually push the button to get it because we can predict that you’re going to do it.
So I think, as it relates to AI and these agents, we want to work with these players. We come from a place of strength because of the unique inventory and the fragmentation of these markets that we’re organizing. We’re going to build our own agents as well. As long as we’re thinking about the consumer, our earners, and our merchants, I think we’ll be okay.
Let’s say I start Nilay’s Model Company with Nilay’s Voice Assistant, and I want to get people sandwiches. I want you to be able to say, “I need to go to the airport, get me a car. Go read my email, figure out when my flight is, schedule a car.” This is the dream. For some reason, we’ve got to throw giant data centers at this problem. This is what they pitch me.
So, I come to you and I say, “I want to be able to get cars from Uber.” What is the percentage toll? What is the extra margin you would have to charge me, in dollars and cents, to make it worth it for me to take the customer away from you in that way? Because when the customer opens your app, you get to cross-sell them into Uber One or ask if they would like some food when they arrive. There’s all these other incremental opportunities that you would forego if I take that customer.
So I have a weird philosophy on this. Initially, I charge you zero. I think that companies sometimes–
Alarm bells just went off at every level.
No, listen! People spend so much time trying to figure out what the economics might be when the first thing is to try it out. Is it going to be a good experience or not? Is your scheduling actually going to work, or is it going to be off and the driver has to wait for 10 minutes, which is terrible? Let’s just figure it out. Then, once you optimize the experience, we can measure. Are you an incremental consumer for Uber or are you totally cannibalistic?
If it’s cannibalistic, then I’m going to charge a lot of money. You can’t have any money because you’re getting the benefit and my content. You’re not bringing me any business at all. If it is incremental, then I would pay some take rate. Is it a 5 percent, 10 percent, 20 percent take rate? It depends on the incrementality.
But I think so much innovation has slowed down because companies try to figure out the economics first. Figure out the experience first and then the economics. Listen, if I do a bad deal for a year, who cares? I’m going to renegotiate with you. I’m building stuff for the next 10 years. Success or failure isn’t going to be determined by my take rate being 5 or 20 percent in year one. It can set precedent, and precedents are dangerous. That’s why I would say to charge zero. Let’s try it out. Let’s see what the experience is. Let’s try to measure out what the value add is, and then the economics will take care of themselves.
You said you’re having these conversations a lot internally at Uber. What is the shape of those conversations? Are there partners you’re excited to work with? Are there agents you’ve seen that are promising?
We’re working with a number of players. I’d say OpenAI is probably the top player we’re working with as it relates to agents. It’s imperfect. It’s really early. The volume is tiny right now. The work is just to make the experience really great. It’s way too early for me to tell you if it’s going to work or not. Eventually it will, but I don’t know the shape right now, so I want to experiment.
When you think about that particular future, everybody wants this product. Again, 60 percent of my conversations are people just dreaming of a future where you talk to the robots and the robots go do stuff for you.
Not a bad future.
That implies a lot of things. At the very beginning, it implies that Uber exists so you even have an API for you to go ask for a car from. At the base layer of that it implies that Toyota exists, and that there are literally vehicles with an effectively standardized service, support, and maintenance network around the world. The Toyota Camry is an atomic unit of transportation across the entire world, and that’s an amazing thing that you depend on so that people can talk to the robot and the car will show up.
There’s another change happening here that I think is fascinating, which is the move to autonomous vehicles and fleet operators for those vehicles. You are a participant in that with Waymo in Atlanta and Austin. Part of your recent announcement here is that VW is going to launch in Los Angeles with you guys next year.
That’s going to change the dynamics of those cars completely. You don’t have independent contractors with the Toyota Camry atomic unit that you don’t have to worry about. Where do we get wheels for a Toyota Camry? It’s just not a problem that Uber has to solve. Cars start driving themselves and suddenly, a lot of people have to worry about different things. How do you think that’s going to change while the agents are coming for you?
Oh my god, now you’ve added agents to the mix.
At the end of the day, I think what you really want to do is say, “I need a car,” and then there’s just a network of autonomous vehicles waiting to accept bids.
I think you’re correlating two things that I would separate. I was trained as an engineer, and you want to separate and simplify the two problems, although they may interact in certain ways. We talked about agents. If I bring value in by creating inventory for that agent, the value’s going to be determined based on whether that demand for the agent is incremental or not. If it’s not incremental, the value’s really small, which means the content provider should get the vast majority of the economics. If it is incremental, then we’ll talk about the value of that incrementality.
Now, when you look at autonomous vehicles, you’re essentially replacing the driver with a machine. Our everyday drivers not only drive. That’s their main job, but they usually own the car, refuel the car, clean the car, etc. So you have to replace that work set. First is the software layer. Waymo is working on it, there’s VW. There are tons of different companies, like Pony and WeRide, that are working on that. We want to work with all the software providers, make sure they’re safe and affordable, and bring them to market.
Second is the car itself. The car can’t just be a good, old-fashioned car. It has to be a super car. It’s wheels but also a sensor kit. It needs much, much more compute to work, and the cost of that hardware right now is inordinately expensive. But it will come down after a couple of generations. Then, you have fleet management, which we just talked about and involves cleaning cars, etc., and then fleet ownership as well. The manager may own the car or ultimately… actually I think you’re going to have car ownership by the Blackstones of the world. Right now, Marriott doesn’t own any of its hotels. There are these entities called REITs that own the hotels. There will be fleets, with these financial companies and retirement pension funds owning big fleets of cars.
Our job is going to be bringing demand to these really expensive cars. These AVs are so expensive, you’re not going to have that many of them on the road in the early days. So, there will be a hybrid model like we have in Austin right now. If a Waymo happens to be close to you, we’ll hail a Waymo. If the Waymo is 10 minutes away and there are a bunch of drivers who are really close, we’ll get you a driver.
The percentage of AVs over a period of time is going to increase, and our job is going to be to manage that shift smoothly and make sure that the utilization, whether it’s a Waymo or VW, is really high. Right now, what we’re seeing in Austin is that the average Waymo is 99 percent more productive in terms of trips per day than the average driver. So we can really drive utilization.
Ultimately, if the demand for that Waymo comes through the Uber app the old-fashioned way, through an Uber agent, or through an OpenAI or another agent, I don’t think it really matters in terms of the autonomous transition. That’s going to happen separately. We want to manage it. Where that demand comes from, whether agent, app, or something that we haven’t imagined, that’s a separate issue. So we just have to manage both transitions. I don’t think the two are really going to conflict with each other.
Let me try to make them conflict.
All right. I like it.
You have some big partners. I think a lot of people assumed Waymo would be a competitor, but you found ways to partner with them.
I think they’re a competitor and a partner. I’ll give you an example. Domino’s is a competitor to Uber Eats because sometimes people go directly to the Domino’s app, and it’s a partner at the same time. I think people look for the drama, like if you are with me or are you against me. The fact is that they work with us — coopetition or whatever you want to call it. Waymo’s going to work with us in certain circumstances, and it’s going to be looking to bring customers directly in others.
Right, because you control a whole bunch of demand for their product.
Yes.
The one competitor I don’t see making as pragmatic a decision as that is Tesla. We will see how [the Robotaxi] actually exists, but the way it has been described is it’s going to take everyone’s individually owned Teslas, flip on some software, and while you’re sleeping, they’re going to go off and drive at night.
I have a lot of questions generally about where the demand in the middle of the night comes from, but so it goes. But that’s a system where you suddenly increase a bunch of autonomous car inventory that is just floating around the city, and you can get demand by just saying, “I need a car.” Then, you have an individual unit of supply that’s like, “Here I am.” You can build an entirely other system that does not require paying you a transaction fee.
That does feel like how you can combine these two ideas because the individual operator of the Tesla Robotaxi might have a different approach. How do you reconcile that? Do you think Tesla’s going to make that work? Do you think that all those people are going to end up turning over their houses to Airbnb management companies like Airbnb owners do?
I think it remains to be seen. We have to take anything that Tesla does seriously. Elon Musk’s an unbelievable entrepreneur. He’s built so many companies. My thinking is that the individual owner of the Tesla is going to get a subset of demand if that demand can only come from Tesla. Then, he or she could get from Tesla demand plus Uber demand.
So 10 years from now, every single Toyota comes with self-driving software. I think that is relatively likely. You can buy a Toyota and when you’re asleep put that Toyota on Uber, which has unbelievable amounts of demand, and make X dollars per day. Or you can buy a Tesla and put it on the Tesla app, which I think is going to have less demand. Ultimately, if you’re looking out for the owner and you want to allow them to monetize as much as possible, that owner’s going to look for the most demand, which is actually going to drive them to multi-source.
This is why five years ago, Domino’s was on its earnings calls saying, “I’m never going to use third-party apps,” and it is now. Ultimately, it’s a restaurant, and you want to monetize that restaurant as best you can, and having multiple platforms feeding that restaurant, so to speak, is mathematically the best way. So, I’m hoping that Tesla eventually decides, “I can build my direct channel, but I can also work for Uber.” That’s ultimately best for Tesla owners.
Have you talked to Tesla about collaborating in any way?
We talk to them all the time. I think there are 150,000 Teslas on the network. We’d love to collaborate with them, and if they want to take us up on that, we’re game.
Do you think they can actually pull off the Robotaxi fleet as they’ve described it?
I don’t know. From my standpoint — and this is judgment — you really need superhuman safety. Superhuman, to me, doesn’t mean better than a human. It means five times better than a human. I think the data suggests that Waymo is around that level. It’s not 100 percent clear to me whether camera-only can get there. Most of the players who are building this technology from the ground up are going with cameras and LiDAR to have redundancy as it relates to perception. If Tesla can do it, more power to them. We’ll see. I think time will tell.
The thing that I think about is what I already mentioned, which is that I know what kind of riders exist in the middle of the night. I was one of those riders in my 20s, and it just seems like there’s going to be an awful lot of puke in an awful lot of Teslas.
There’s a lot that drivers do that we take for granted. There’s a lot of work.
How do you manage that with your autonomous fleet now? Somebody gets sick in the car, somebody makes a mess in the car, and you don’t know about it until the next rider shows up.
That’s something that you have to handle, which is a rider complaint, or you might have cameras inside the car, for example. The newer cars have that.
Do you make an AI puke detector?
I guess we’re going to have to build that. That’s going to go into our product roadmap.
Smell-O-Vision in the car to make sure.
That’s an interesting one, but that essentially is going to be the job of the fleet operator.
How does that cut work with the fleet operator? I took an Uber to work today. I figured I should use your app before I talked to you. I asked the driver and he said, “I want my rates to go up.” Maybe every driver I’ve ever talked to wants that.
Everyone wants to make more money. Absolutely.
That’s the driver’s perspective. I’m curious how you’re going to make the rates go up with big fleet operators who now have giant fixed costs, are happy to have 24/7 utilization, and maybe take a lower rate for greater use while the drivers just want high rates. How do you square that circle? It feels like the driver rates are going to get pushed down as autonomy comes into the mix.
Our job is going to be managing the balance between demand and supply. So, I think autonomy is actually going to drive more demand. Even though autonomous inventory is going to be an ever-increasing percentage of our overall inventory, the number of drivers that we need on our network over the next 5-10 years is going to increase because demand increases more, so to speak.
Twenty years from now, you might be right. From that standpoint, we just have to manage our inventory. There’s a turnover of drivers all the time, and we want to make sure that we give drivers the right earnings expectations. In a city where demand isn’t growing because autonomous supply is coming in, maybe we just have to slow down driver recruitment and not make them promises as it relates to earnings.
So, I think communication, being straight with people, is the ultimate answer. But I don’t take it for granted. This is going to be a transition that we have to manage very, very carefully and make sure that our driver base is taken care of.
What do you think the deals with the fleet operators will look like? Maybe the entire thesis of The Verge is that adding computers to things makes them very complicated, and we should pay attention to that complexity. I’ll use the Toyota Camry again, but with the Honda Accord or the Chevy Malibu my parents had in the ’80s when I was a kid, you didn’t have to update the software. They didn’t have buffer overflows. The infotainment deck did not restart in the middle of the day.
They might have cranky mechanical issues, but again, there’s an ecosystem to help fix that very quickly. If you get a flat tire in New York, you can just push the car around the corner and there’s a taxi garage waiting for you. That’s incredible. You’ve got a bunch of cars that have tons and tons of software at the bleeding edge of technology, yet cameras can get dirty or broken or whatever. That’s a different set of costs on top of the existing mechanical costs. You’re not going to repair the robotic cars. The fleet managers are going to.
We actually work with fleet managers today, and the reason is that a bunch of drivers can’t afford vehicles or can’t get loans for vehicles. Companies can. Also, we’re actively trying to electrify our vehicle base. Close to 10 percent of our vehicles are now electric, with more than 20 percent in certain cities in Europe. So, we have experience with fleet operators. About 15 percent of our volume comes from fleet operators. Many of those fleet operators own the cars. Not all fleet operators own the cars. I think if they own the cars, then they’re just looking for a return on invested capital, and the return is going to depend on the risk. The more volume we bring them, then the cheaper the cars are going to be because utilization is going to be really, really high.
Then, there are going to be some pure play fleet operators where the financial players are going to own the cars. That’s going to look more like cost plus. So if it costs you $10 to operate the fleet, we’ll pay you $11 or $12. Our philosophy right now, as I said earlier, is to make the damn thing work, make it safe, and then worry about the economics. AV loses money for us right now, but we know it’s going to be huge in the future. It’s going to ultimately bring the cost of our service down, and it’s going to make the street safer. For us now, it’s about making the system work, and then the economics are going to fall into place.
But I think fleet operations will be a part of the ecosystem and they won’t be the majority. What we will bring to these autonomous players is an end-to-end solution. So, they have to work on what they’re really good at, which is self-driving algorithms, and then leave the rest up to us.
But it’s that middle part that I’m just focused on.
Well, we work with these fleet operators. For example, one of our fleet partners is operating the Waymo fleet in Austin right now. Certain players may want to do the fleet operation themselves, but I think that our local on-the-ground expertise and the fact that we are running fleets with millions of cars today will be another benefit. I don’t know if VW in LA is going to want to operate its own fleet. We will be able to bring that capability. If they want to do it themselves, that’s totally fine as well.
I think I’m just stuck on that the fleet operators are going to have a bunch of really highly paid IT people sitting around rebooting the fiber optic networks of these cars.
If it’s just the reboot, that’ll be easy. Again, these are complex vehicles. The cost of the vehicles is really high right now, and the care for those vehicles is going to be not-zero.
When you think about the safety side, which you’ve talked about several times now, the question of autonomous liability exists in the world. I book a Waymo through Uber. The Waymo gets in an accident. Who accepts that liability today?
We think early on, most of the self-driving companies will self-insure. They’re big companies. They can afford to self-insure. But ultimately, whether they insure, we insure, or the owners insure, it’ll work itself out.
What is the mechanism for working it out?
Looking at the incident rate and looking at the cost per incident. I think incident rates are going to come way down, especially if these autonomous vehicles are superhuman safe. Cost per incident is going to go up, which is because the cars are much more expensive. But I think that the overall cost of insurance as a percentage of your ride is going to come way down because of the safety of these cars.
We’re almost out of time here. I’ve got to ask you the last question in honor of my Uber driver today. If you were to tell Uber drivers today, “Here’s how I will make you make more money,” what would you say?
I would say we’re going to bring you more rides. The income for Uber drivers in New York, for example, is more than $50 per utilized hour. New York’s an expensive place, but we think that’s a pretty good trade-off for a job that brings lots and lots of flexibility. But it’s a hard job, traffic is tough. So, we’ll keep working on it. Ultimately, the way to offer more earnings to drivers is to bring more demand, and every single day we’re working on that.
Dara, this is great. Thank you so much for being on Decoder.
Thank you. I really appreciate it.
Questions or comments about this episode? Hit us up at decoder@theverge.com. We really do read every email!