Reid Hoffman | Aurora Reflections

Foreign.

Hi, I'm Chris Yeh, the co author of Blitzscaling and I'm here once again with my co author and old friend Reid Hoffman, the co founder of LinkedIn and an investor at Greylock Partners. Today, Reid, I'll be asking you to reflect on your investment in Aurora, one of the leading autonomous vehicle startups. In fact, you've invested in Aurora multiple times since you led Greylock's investment. And now Aurora is going public by merging with a re invent technology partner, spac. When this transaction goes through, Aurora will raise up to 2 billion doll and be valued up to $11 billion, which is pretty impressive. So let's start with the beginning. Describe meeting with Chris Urmson before he started Aurora. What insights did he have then that led you to stay in touch with him?

So Chris, great to see you in person. This is the first time in 18 months that we've been doing this. We've been doing this remotely for the last year and a half and this is a hope to a slowly coming back together world. Now to answer your question, I saw Chris Urmson first. I met him at Sun Valley where he was presenting the Google self driving project. And it made me realize that the time for this technology was now. That the fact that a whole set of challenges which had seemed remotely infinite from that there were just so many problems you had to solve in order to get autonomous vehicle right that Chris, who was the technical lead, who'd solved all the different problems in terms of how to look at, you know, kind of sensors, how to look at planning, how to look at this infinitely long tail of different circumstances on the road and to pull all that together, that this future was now here. This wasn't a science project, it wasn't a Xerox PARC project, it was a we are building this and this future will be here. You now have line of sight. And so Chris and I started spending time together advising him on the various things he was doing. And then when he decided that it was better off leaving Google, and then we had initially left Google and then said I've got an idea for what happens if you start from scratch and rebuild from the foundations up. I was honored to be amongst the first phone calls that Chris made after of course having, you know, spent time on this with Ian.

Got it. So when was that first phone call? When did he reach out to you?

Well, as you know Chris, my brain is not really a historian's brain. I don't have a like, you know, it was February 13th on a dark and Stormy night or Buck's breakfast at so and so. I do remember that we did meet at Bucks because it's kind of that iconic place of Americana within Silicon Valley. And we were talking through and it was kind of like, okay, well the big question obviously was, well, so you spent a bunch of time building it, making it work at Google, you've got it working now you're going to be starting entirely from scratch. Given that these things are, as per Chris, you're in my work, a race to scale in blitzscaling. What makes you think that this is a race that since the metaphor, the other race car is far down the track and is driving and you're going to go build a new race car and overtake it, how does that work? Were obviously a bunch of the first conversations and I was persuaded that the stack of things that Chris was thinking about and, and it arranges everything from technology to business models to how do you become the first to scale or at least be in the contention set that he had thought through all of these things and that this was a worthwhile startup bet because he understood the landscape, he understood the cars, he understood the driving, he understood the terrain and what was the thing to do. Kind of mixing metaphors as it were.

So what we do know is that you made the investment in Aurora, led their series A in 2018. So classically speaking, we've talked a lot about the importance of investment thesis and also timing. So what made 2018 the right time to invest? What was your thesis for investing?

So part of the thing, if you would try to do this much earlier, it's one of the benefits that large tech companies have is they can invest for a long time without a foreseeable future. When you do in venture, you actually are really making a prediction about availability of the market, was that you could see line of sight to the technology, you could see what circumstances work, you could align business models, you could align what you saw with what was going to happen with all of the large scale OEMs. And this one was a little bit there's some risk, you were taking some risk on what does regulation look like. But you had reason to believe the States, federal government, other countries would be making the right bets when it came to regulation. And so 2018 gave you the window by which build deployment, be the scale technology provider for this amazing technology could all happen within a venture time frame.

The other thing you mentioned of course is that Chris came from Google and there was this race car that was already way ahead on the track and Google's not the only company that's tackling autonomous vehicles. Everyone seems to be into it. How did you know that Chris and his founding team were the right people to execute on this idea and shoot to the front of the pack?

So one of the things that, with a kind of a tongue in cheek that I like to do is I like to call Chris the Henry Ford of autonomous vehicles, because he's the only person who's actually, in fact, done it so far. Because, you know, Waymo has cars that are autonomously driving in suburbs of Phoenix. They were built technologically, you know, under kind of Chris's guidance and build, and so he knows what the entire roadmap looks like. You know, obviously some of this I won't be able to go in depth in because some of it's, you know, kind of the trade secrets of, of Aurora. But what Chris had thought is like, okay, now that I've done this once, now, if I was redoing it, what fundamental technology platforms would I do differently? What things that is now present that wasn't present when I was first doing it? Some of that is how do you make simulation work really, really well? How do you make simulation work for the long tail of challenging cases? So you're amplifying all of your driver miles into a much deeper learning set, which obviously will matter as you have vehicles on the road actually deployed and what your continual learning curve looks like. And also key things around, like sensors, what kinds of sensors you would need, what kinds of environments they would need to work in, what kind of visibility you would need to have on the road. You know, these kinds of things done the right way with the right kinds of sensor fusion and then together with a business model. Because, you know, like, for example, one of the things that sometimes debated is how many sensors do you want on autonomous vehicles? And the short answer is you want them all. You'd like to have your autonomous vehicles have as much visibility as possible to help make it as safe as possible and help communicate with each other. And so which of these sensors would you bring together that would be really key for that? And then if you have a business model that isn't, I'm just selling the vehicles, but as a. They're in a sharing fleet, whether it's, you know, kind of like Uber or whether it's a trucking and that's working, then you actually, in fact, you don't have to have a cheaper balm to it. You have all the sensors that are there and they're integrated into a much more Effective autonomous vehicle, much safer autonomous vehicle pattern to make it work. And that was all part of Chris's thinking about kind of 2018 now. He also then of course brought in Drew Bagnell, shared Carnegie Mellon experience, Uber experience and Sterling with Tesla experience. So you're not only getting the in depth Google and its chauffeur self driving project to actually deploying, but also this depth of expertise from these other major efforts to say if we're rebuilding from scratch, what are the things we know now to have the right scalable technology, the right safety technology and the right business model to pull it all off.

And what I like about that classic sort of judo strategy approach is that it's something that a startup can pull off where you have these established competitors that have already been building and building for years and years, they've gotten approach, it's really hard for them to start from scratch. And so by choosing this approach, Chris was able to find a way to really leverage the strengths of being a startup.

Yes, exactly. And part of it was then to also say let's recruit like Chris already had an entire map as to the best talent was let's recruit talent from everywhere, including universities in Canada and other kinds of places as a way of building it up. We know how to structure the project from the very beginning because we know what the decomposition needs to be. What does the hardware side, what is the sensor side, what is the software side. We know like through all the trial and error that had happened at Chauffeur, like I learned from Chris, what were the code names when I talked to other self driving projects that came to pitch me for financing. So how are you going to handle highway construction? So how are you going to handle long tunnels and a bunch of other things? Because they said oh yeah, it'll just be the same generality. Ah, you don't realize that these are special problems because you haven't encountered them yet, which means you're still not, you know, very deep and far along the race. And so those kinds of things, Chris was unique for pulling that all together and the startup allowed you to do it essentially from a fresh foundation.

So set the stage for us going back to 2018. What was the state of autonomous driving in vehicles at the time time and how far has the field come since then?

So you know, one of the things, as you mentioned earlier, there's lots of startups, I do think that it's only a small number of these startups that actually in fact have all of the experience that comes from having done it already. Versus the oh well look, we're just applying modern computer vision and some new things to this problem. It's how are you actually in fact, you know, making this actually in fact all of the stuff work because it's a multi component problem as a way of doing it and they've done it before. Now Initially back in 2018 all of the OEMs were panicked, thought that they would have to do it externally, were knocking on the doors of all of these startups. Some acquisitions happened from that, some initiation. Now I think the OEMs broadly think oh yeah, this is going to take a while to get there so we can all build it ourselves. I think that pendulum is likely to swing back because actually in fact as the question is, look, Google already has some self driving going in Phoenix, it's here, it's just a question of how that gets deployed within the OEMs fleets and scaling and you know, Aurora has been making the various moves that it is thought would be critical to make that happen. So I think now the OEMs are thinking it's too long where it's actually in fact nearer in.

One of the classic pieces of advice you give to entrepreneurs is to ask your smart friends what might go wrong with their ideas. Now when it came to Aurora, what were your critiques of the early ideas? What pushback did you get from other investors and ultimately how did you convince them to invest?

So with Aurora there were probably four primary questions which we are obviously still working through but have made good progress on all four. One is can you deliver on the technology in the timeframe that you need to have operational product? Another one is kind of will government regulation allow the certification, allow the safety, allow the deployment of it? Another one is will you have a business model that will sustain the value of this kind of commercial investment that will also have that compounding loop of accelerating the deployment of your technology? And then of course can you get to the scale delivery of this? And part of just on some touch points which we will obviously get into some depth, part of what Aurora has done between 2018 and now is, has built out a whole bunch of the technology done, you know, validated some parts of its sensors and its planning and its simulation and other kinds of things, has worked with the regulators to establish what kind of thing would be good across all OEMs as well. Doing that, has been working on it, establishing the right relationships to have the kind of business model. And all of this leads to getting the car assembled to drive down the road towards scale delivery.

What are Some of the smart choices that Aurora's made along the way that have helped get the company to the position it is today.

A bunch of the stuff the company has already kind of talked about in different pieces. They have made some extraordinary sensor acquisitions because as mentioned before, when you say, well, what does it really take to deliver a high safety vehicle when there'll be a number of these vehicles on the road, sharing them with other human drivers, what kinds of visibility do you need from the vehicles? How do they need to operate in order to do that? So there's a bunch of different sensor technology primarily through acquisitions, where it's both the raw technology itself, but also the teams that then know how to. Because part of Chris's vision along with his team was how do you then refocus that technology on the autonomous case for these circumstances, since we know a lot about the autonomous case from the work that they've been doing before. And how do you make that happen? The simulation Aurora built a lot of its own simulation that was trying to make every mile driven. Because like it isn't that you've just driven like let's take a parallel from the Malcolm Gladwell. It says that you just had 10,000 hours. It just not you had 10,000 miles. It's what do you learn from the miles? What's your learning coefficient? And to use simulation as an amplifier to make that work and have that loop done the right way. And then of course obviously it was a massive combination with ATG from Uber because when you know Uber was making decision to saying, look, we should do best to breed on this, we should like what matters to Uber is not that it owns its own unit, it has the best technology for deployment at scale within the Uber fleet. Then the combination was, okay, we'll select the Aurora technology to combine these great technological assets and capabilities. But also part of the value for Aurora in doing this was then you then have a scale deployment path to all of the Uber services around there and that convinced them to then reprioritize. In addition to trucking, which Aurora had been prioritizing and it had a number of great partnerships that have already been announced within the. The trucking arena had also said, okay, well we now also then have scale passenger cars as well. And that that combines to being that.

Scale delivery solution, it's easy to look at a company like Aurora that's going public and think it had an easy road to success. After all, we have this guy Chris who helped build self driving vehicles at Google. He's the Henry Ford. But of Course, things are never that easy. So what kinds of major challenges and difficult decisions did the team face along the way?

Structurally, the top one was speed to market with a complete safety focus. Because one of the missions within Aurora is to save tons of lives. We lose lots of lives through road accidents and many more than, like, people more worried about flying. But you're more likely to lose your life driving to the airport than you are on the plane flying in terms of statistics. And that's of course, partially because driving is very dangerous. And how do you use autonomous vehicles to make it more safe? So on one hand you go, well, this is a speed to launch, a speed to getting your product deployed. But on the other hand, at complete safety, and that's the macro challenge that they've been navigating. Now, within that, they kind of start with the kind of focus because this was kind of the, you know, the problem that Chris had been solving on passenger vehicles. Then they pivoted and really focused on their truck relationships and the truck development. Because while the platform is still the same, it's tuning differently, establish all kinds of partnerships. And then perhaps one of the biggest, you know, decisions or I had to make is in the Uber deal was to also then say, okay, we had been super focused on trucks and now we're adding concurrently passenger cars back into it. Because that's one of the things that matters to Uber. We're doing this. And those are kind of giving some highlight of some of the strategic decisioning choices. Because you start with passenger cars, you go to trucks, and then you come back and say, okay, well, passenger cars, we knew we wanted to get back to them anyway. We'll also add that in because the acceleration that we get on the overall business, the overall technology, the overall scale deployment with the Uber deal is such that we will do trucks, trucks, trucks, trucks and passenger cars.

Well, you've mentioned a couple times the Uber deal, and there were certainly a number of things like this important moves that Aurora made along the way, whether it was launching a particular product, partnering with a particular company, even some of the M and A activity that helped the team get to where it is today. What were some of those really important things they did?

So we've talked already about sensors and Uber, so I won't rechat on those that much. But what I will add in is, and I guess I have to be a little vague here only because I want to make sure that I'm not adding any new information that isn't already in the various public disclosures that Aurora has Made in its combination with the Re Invent technology partners back and going out. But it has got a number of key trucking partnerships for development R and D that if you look at the number of trucks manufactured per year, these partners account for over half of the truck manufacturing. They've made a number of specific deployments and development work in this trucking industry. Look, we have a massive trucking driver shortage. So there is a huge need for autonomous trucks. That has nothing to do with the changing landscape of jobs because like literally it's one of the most needed thing is having a whole bunch more trucks to make logistics work. And so all of that are among the key moves that Aurora has made to where it is now.

We've spent a lot of time talking about some of the changes that have happened along the way. What are some of the core aspects of the company that you saw in the beginning that are still evident today, what stayed the same and what has helped the company get to where it is.

One of the reasons why when you look at investing, you tend to be on the series A, the seed is you tend to say like people tend to say, well, is it market, is it the strategy plan or is it talent? And by the way, it's all always, it's not an either or. But you have a huge emphasis on talent. And part of the reason you have a huge emphasis on talent is because frequently in the C, in the series A, things will change. You'll pivot. There will be unknown challenges that you'll have to deal with in order to get there. And so, for example, as part of Aurora is like, when we're sorting through it, we said, okay, we'll start with passenger cars because we know how to do it. Election tech, the business model and getting the trucking stuff will be a higher priority. So we'll shift to trucking as a high focus. And then as part of doing trucking and high focus, you go, okay, we've been doing that. Oh, look, and we know we're going to get back to passenger models, but we're going to do it because Uber is now coming and saying, hey, this is a huge opportunity. And so you, you are shifting to the opportunity that's in front of you. And that's part of the reason why. Talent, talent, talent. And so that was part of a major part of what we got right, what was really important. And the strategy was right. But the evolution and change of it was a changing circumstances in the market, changing circumstances in what the OEMs were prioritizing, changing circumstances on where did the scale business model look like? And those are where you get to changes as you drive down the road, right where you're kind of saying, okay, which path do you take to this autonomous vehicle future? And so I think that the really key for getting all of that adaptation as we've been going from 2018 to now was talent.

I do like the way that you were able to go meta there to be able to use the metaphor of driving as a description for the journey of figuring out autonomous vehicles and, and self driving cars. So well done, well done.

Thank you.

So Aurora believes that in the end the autonomous vehicle space will be dominated by a small number of players. What about Aurora leads you to believe that it can beat those many competitors that are out there and become one of those few enduring market leaders.

So I think Aurora has a very good shot at being the first to scale tech. Obviously Google's had important tech there for a while. There's obviously other players in the market that are describing why they think they can be. I think that the questions for Aurora are things we talked about, which is by having done it before, they actually have a very good sense of what do the sensors need to be like, what does the simulation need like, what does the learning per mile need to look like? What is the business model there? Why trucking is a primary focus in going in together given the melee that there is around passenger cars, say. Well, actually, in fact, doing this with the scale sharing model of a network like Uber is a very key way of doing that. And I think that's that set of like it's this massively huge project with all of these essential elements across them and making them all work together to deliver within a compact time frame is part of what I think the magic of what Aurora has been pulling off.

Excellent. Well, Reid, thank you as always for taking time from your busy schedule. It's great to be together in the same room recording. Hopefully people can tell that from listening. That concludes this episode of Gray Matter. You can subscribe to our podcast on soundcloud.com greylock partners and you can also find new episodes and blog posts on our website@greylock.com. you can also follow us on Twitter at greylockvc. I'm Chris Yeh and on behalf of Reid Hoffman, thanks for listening.