Five years ago, venture investors, tech companies and automakers were pouring tens of billions of dollars into driverless cars. Tesla, General Motors, Lyft, Uber, and Google’s Waymo were promising large fleets of robotaxis with fully autonomous vehicles by the turn of the decade.

In 2017, Ford took a big swing. The company invested $1 billion in Argo AI, a startup developing level-four driverless systems. Later, VW entered the partnership. The automakers promised to make a fully autonomous car by 2021.

But in October of last year, VW pulled out of the partnership. Ford said it would shut down the driverless car program, taking a $2.7 billion loss.

So how did we get to a point where a promising startup valued at $7 billion is being written off by automakers? And what does it say about the viability of fully-autonomous cars?

Journalist Ed Neidermeyer says Ford’s shutdown of Argo AI was due to inflated expectations – which exposed a mismatch in business models.

“I think it’s very easy to look at this and say, ‘shutting down Argo AI was an admission that this technology doesn’t work…or was a scam. And you look out on social media and people are taking that lesson away – and I think that’s the wrong lesson.”

This week, we speak with Ed about the real lessons behind the setbacks for autonomous cars: the mismatch between our fantasies and the reality of the technology.

Full edited transcript:

Stephen Lacey  00:07

There’s this very famous chart in tech and investor circles called the Gartner Hype Cycle. You might be familiar with it. It illustrates the phases of how technologies get adopted. It looks like a sound wave, or maybe a heart rate stretching out over time. At the start of the wave is a technology trigger, and then it rises up quickly. At the top of the wave is the peak of inflated expectations. This is when the money pours in and everyone is promising the moon. And then that peak drops abruptly into the trough of disillusionment. That’s when everyone realizes that those promises are not going to materialize, or they’re way off. And then it works its way back upward, more gradually to the slope of enlightenment, and finally flattens out to the plateau of productivity. And a lot of people would argue we are squarely in the middle of the trough of disillusionment for autonomous vehicles. So that begs the question, Where was the peak? If we want to pick a moment that represents the peak of inflated expectations, we could probably go back to 2017. At that point, venture investors, tech companies, and car makers had poured tens of billions of dollars into driverless cars. Tesla, General Motors, Lyft, Uber, Google’s Waymo. They were all promising large fleets of robo-taxis with fully autonomous vehicles by the turn of the decade. Details about Apple’s secret car plans were leaking out as well. And into that environment steps forward with a big swing. It was February 2017, then Ford CEO, Mark Fields was sitting on stage at a press event with three other executives. And he made a bold proclamation.

Mark Fields  01:37

“We think that automation is going to define the automobile in the next decade.”

Stephen Lacey  01:43

Automation, he said would be as fundamental to the auto business and society as the moving assembly line that made mass car production possible.

Mark Fields  01:51

“And that’s why we’re investing $1 billion over five years, in the next five years in Argo AI, an artificial intelligence company. And their role is going to be to develop the virtual driver system for autonomous vehicles. And of course, it remains our intention to introduce a fully autonomous level for vehicle into the marketplace in 2021 in mobility services.”

Stephen Lacey  02:19

Next to fields sat a young engineer and self driving wunderkind Brian Salesky, the CEO of Argo AI. He expressed confidence that level four autonomy, a car that can fully drive itself under specific conditions, was around the corner.

Brian Salesky  02:32

We want to move with urgency. The business structure is very unique in that we have the agility and speed of a startup but combined with the product development, and scale of Ford. We think that that’s what’s going to be needed to bring this important technology to millions of people.

Ed Niedermeyer  02:48

Argo AI was the startup that was created by a gentleman called Brian Salesky. He’d been part of the Google self-driving car project before that became Waymo. And before that, he was in the DARPA challenges with Carnegie Mellon University. He also worked with Caterpillar automating the mining trucks a decade ago.

Stephen Lacey  03:06

Journalist Ed Niedermeyer followed the big unveiling from Ford and all the other activities swirling in the autonomous car space. It was a moment when automakers were stepping in to help commercialize driverless systems, partly as an innovation play, partly out of fear.

Ed Niedermeyer  03:20

Venture capitalists put a certain amount of money to create these companies to develop this this level-four technology. They needed more money and the venture capital sort of passed off that investment to the automaker saying, “Look, this technology is going to happen. You’re either part of it and you have a new business, or you get disrupted by these guys.” And automakers bought that and so the automakers are behind the wheel.

Stephen Lacey  03:40

In 2019, Volkswagen entered a joint venture with Ford and Argo AI, Argo was valued at more than $7 billion. From the outside it was a bet of confidence in the technology. On the inside, however, the partnership with Argo AI was starting to swerve off course.

TV Host  03:56

Let’s now talk about cars, specifically AI cars. The artificial intelligence car market seems to be crashing. Difficult day for many in the tech industry is hundreds of employees at Argo AI found out they’re losing their jobs. Two have the self-driving car company’s biggest investors decided to pull out. The company landed here six years ago thanks to a multibillion-dollar investment from Ford and Volkswagen, but now they say they don’t see a path to profitability on fully autonomous vehicles.

Stephen Lacey  04:26

In October of last year VW pulled out of the partnership, Ford said it would shut down the driverless car program led by Argo AI, taking a $2.7 billion loss. At that point, Uber had sold off its self-driving unit, Waymo and GM’s driverless taxis were still publicly having learning problems on the streets. Tesla was facing a criminal probe from the Department of Justice for inflating autopilot capabilities, and a handful of startups were facing cashflow crunches. Niedermeyer says Ford’s shutdown of Argo AI was due to inflated expectations, which then exposed a mismatch in business models when things weren’t going right,

Ed Niedermeyer  05:01

We hear the terms autonomous vehicles, self-driving cars, and we think people are developing cars that we can own and that can drive themselves. What Argo and Aurora and Waymo and Cruise and the other major serious players and autonomous vehicle technology, what they’re actually building…they look like cars, right? They’re four-wheel vehicles, they’re about the size of cars in a lot of cases, and they’re currently regulated as cars because we don’t have a different structure. But fundamentally, they’re very different because these are not vehicles that we’re going to own.

Stephen Lacey  05:31

Level four vehicles are really expensive. They are not meant to be mass produced. At current costs, they could only work in fleets for high volumes of passengers, not exactly the core business of automakers. When it looked like level four autonomy might be realized quickly, automakers like Ford and VW were willing to spend a lot of resources. But as the technical and financial challenges mounted, the sheen wore off.

Ed Niedermeyer  05:54

The result of this was that when the car companies ended up at the financial steering wheel of these level-four autonomous vehicle companies, once the hype started to subside, I think the conclusion was simple. It was this is fundamentally different than the car business and we don’t really have any good reason to invest in it.

Stephen Lacey  06:14

With the end of Argo AI, a lot of people are saying investments in autonomous vehicles have been a failure. So have they?

Ed Niedermeyer  06:20

It’s a big deal. And I think it’s very easy to look at this and say shutting down Argo AI was an admission that this technology doesn’t work or wasn’t going to work or was a scam and you look out on social media and people are taking that lesson away. And I think that’s the wrong lesson.

Stephen Lacey  06:38

This is The Carbon Copy. I’m Stephen Lacey. This week, we’ll explore the real lesson from the technical troubles in autonomous driving: the bubble has popped. So what’s next? Ed Niedermeyer is a journalist, columnist and author of the book Ludicrous, The Unvarnished Story of Tesla Motors. He’s been covering the automotive space since 2008. In the autonomous driving space for nearly as long as a decade ago, he got to drive in a car operated by Google’s Waymo and it totally shifted his perspective.

Ed Niedermeyer  07:17

It wasn’t just this abstract technology. This car, this vehicle, it really wasn’t a car. It was a vision of something that was beyond the car. And I think the idea of driving automation technology as something that could take us to a world beyond cars, was something that was really interesting to me. And so I just couldn’t resist the siren’s call.

Stephen Lacey  07:37

At what point did it feel like the expectations for where autonomous vehicles could go was diverging from the reality of technology?

Ed Niedermeyer  07:48

Almost from the very beginning, I noticed there was this massive disconnect between people’s perceptions of autonomous vehicles and driving automation tech, on the one hand, and the reality of the technology.

Stephen Lacey  08:03

Ed was impressed, but the driving was inhuman, it wasn’t naturalistic, the technology just didn’t quite feel ready.

Ed Niedermeyer  08:10

And there were all kinds of very strange behaviors that really leapt out as “we’re at the early stages of this technology.” I mean, you’d be driving down the street, this is in a residential neighborhood in the Bay Area. And there’d be cars that are parked irregularly on either side of the road. And this vehicle would be going basically down the middle of this quiet residential road. And it was like weaving very slightly as if these cars parked on either side of the road were going to jump out at it or something were obstacles. And it was very hesitant making turns, very slow. I was surprised by how, in ways unimpressive, how crude at that point that technology still was. And yet at this at this moment was probably when the hype around this technology was the greatest.

Stephen Lacey  08:56

But the technology was just good enough to capture wide attention. Investment ramped up quickly, promises got bolder, the hype cycle accelerated. Fast forward about five years, an Uber crash, missed milestones, confused cars, they all started to challenge the narrative that robot cars would soon take over the streets. But when Ed rode in a car from Waymo, again, the technology had improved dramatically. There was a disconnect between the fantasy of near-term autonomy and actual progress.

Ed Niedermeyer  09:24

All of a sudden, now these computers were driving naturalistically. They were making crisp left-hand turns, they were moving with traffic, they weren’t holding it up. I could have closed my eyes and I would not have known that I was not in a human-driven vehicle. So the hype was highest when the technology was if you experienced it in person was just obviously not ready. And yet as the technology got more and more impressive, even to the point where I’ve been on a fully driverless ride on public roads in Arizona in a Waymo which is, again, like probably the most miraculous technological experience I’ve ever had. And yet at the time that that happened, people were more disillusioned and skeptical of autonomous vehicles, then at any other point since I’ve started paying attention to them. So we have two things going on here. One, is that the technology is developing and the progress there has really been remarkable. And on the other hand we humans have had this roller coaster relationship with it that exists entirely in our minds, based on just very poorly informed perceptions of it. And so we create the psychodrama that really has nothing to do with what’s actually going on. I think progress in the technology is genuinely impressive.

Stephen Lacey  10:33

How big of a deal is the shutdown and subsequent sell-off of the component parts of Argo AI? How big of a deal is that for this space, both in terms of investment flows and our perception of the technology? So what you’re saying here is that it wasn’t necessarily a technology failure, it was a business mismatch.  Is that correct?

Ed Niedermeyer  10:48

It’s a very big deal. But I think, again, it’s really important to take the right lessons from things. One of the most important lessons from it is that the venture capitalists got a lot of people very excited about this vision of what autonomous vehicles could be, which was different than cars. And yet they also sold these companies, passed their bags, so to speak, to the car companies. And I think that’s why people say, you know, “Well, it’s a failure of the technology,” because if the technology had to be so good that it would actually disrupt these car companies, which is certainly why they invested or one of the reasons they invested. The technology that exists today is not up to the task of automating general purpose driving like we do with cars. We can’t automate, the way we drive cars, because it is a multiplicity of tasks, it’s in a multiplicity of domains. It’s just, it’s just way too much. And so the technology certainly failed, if that was the expectation. And it makes perfect sense that again, that car companies are not going to continue to invest a billion dollars a year to eventually get to the point where we can have these robust ecosystems of robotic mobility, infrastructure, whether it’s improving public transportation, or shuttles or delivery or semi trucks and logistics. And there’s a lot of applications for them but it’s not aligned with the with the car company. So I think it’s important to understand that the car companies are the ones who were pulling the plug here, right? They got put in the driver’s seat of this technology, they’re pulling the plug, the technology failed to meet their expectations, which were unrealistic. And what the technology can do usefully is not well-aligned with the current car business. Yes, it was much more of a failure of humans than of technology, let’s put it that way. I think one of the one of the fundamental problems that that mistakes that humans made about the technology was thinking that it could be generalized. And I think this is something that when venture capitalists invest in companies, they usually aren’t assessing each business as a business or as a path to a business and saying, “Okay, we have a very clear path here. I see they’re definitely going to get to sustainable cash flow here. And that’s what matters to me is the certainty that this company is going to grow into a self-sustaining business quickly.” That can matter to venture capitalists, but especially the last 10 years ago, when the when the money was flowing, that didn’t matter so much as making these longshot bets on things that could be these giant winners, right? And that’s how we got things like Uber. Uber couldn’t pitch “Well, we’ll work hard, and eventually we’ll take over most of the taxi business.” No, they had to disrupt private car ownership as we know it. And venture capitalists need that, the giant pot of gold at the end of the rainbow to bet on. And in the case of autonomous vehicles, the pot of gold was that this technology can be generalized. And the reality of automation of all kinds is that you can’t automate general-purpose tasks. We do not have humanoid robots walking around, just taking care of random tasks for us. If you want to automate something, you have to pick a specific task. And you have to automate that task. So when we think self-driving car, we’re fundamentally talking about automating the ability to drive anywhere and everywhere. And that is the equivalent of a C3PO robot, a robot that’s a humanoid that can walk around, that can talk, that can socialize, that can take care of general tasks. Maybe it’s only a high school level, or a middle school level or whatever of intelligence. But it’s that same generalizability. Driving automation technology is like robots, not like cars in this way. So the fact that we don’t have C3PO walking around right now doesn’t mean that robots are scams, or that they don’t exist. There are millions of robots around the world generating millions of dollars of economic value right now, but they’re simple. They’re an arm with a tool on the end of it. They automate a simple task in a constrained domain, right? And that’s how automation works. Automation is a spectrum that starts with simple tasks and small constrained domains. And the task get more complicated as the technology improves, and the domain gets bigger, right? But cars are fundamentally not a one task tool. Cars are fundamentally a tool for every kind of mobility we need. And so if you want autonomous vehicles, you have to get away from cars; they’re not cars anymore. All of a sudden, autonomous vehicles are, I think, for me, the best way to think of them is as like robotic mobility infrastructure. A car is something that you own to take you anywhere you want to go. It’s like magic, which again, which is why we have such magical thinking about it, right. But automation is much more suited to like a bus, right, because the bus drives the same route over and over again, that’s a simple task, it only operates in one domain. That’s a constrained domain, and you can start to train artificial intelligence on the relatively limited number of things that that bus is going to see driving that route over and over a couple of routes over and over again and get a very high confidence that the AI has trained with enough data, that it’s going to respond appropriately to anything it’s reasonably going to see on that route. That is a very different task than training AI to deal with anything that anyone ever sees in a car anywhere they go in a car. These are, if you understand AI and how creating datasets, representative data sets, I mean, these are not even in the same galaxies. And so I think if you want to move forward, I think what we have to do to move forward in this technology is to understand that cars are one thing and that mobility robots, robotic mobility infrastructure, level-four autonomous vehicles, whatever you would have, we need to come up with a good way to talk about this stuff, right? Because we need to make this distinction. I think this is one of the lessons we have to take away from this bubble and the bursting of this bubble is that a fundamental confusion of different things was at the heart of this and we need to start developing language that helps us separate we know even what are we even talking about when we’re saying words like self-driving car?

Stephen Lacey  17:13

What is driving that roller coaster in that the perception difference? Is the way the press covers the technology? Is investor hype and expectations? What is the mismatch?

Ed Niedermeyer  17:25

The AV hype bubble is partly a product of Americans trying to understand this technology as consumers. The last frontier in the evolution of the car is often seen as the ability for the car to drive itself, free us from the driving task. And so I think part of the hype was as consumers, this is something that we’ve been waiting for a century. The other piece of this is the sort of investor perspective and the kind of financial speculative bubble that surrounds technology. You know, venture capitalists like to pose as these rugged individualists who surveyed the emerging tech landscape and blaze a bold individualistic path through it. But in reality this is an area of endeavor that is extremely governed by herd dynamics and trends. And, and so I think one of the things that happened was Google was very quietly working on this technology for a while, at a time when Google was the biggest player in Silicon Valley and high tech. And one day, they said, “Well, here’s some stuff that we’ve been doing with this. Here, look at video of a car driving itself, look at this vision of a totally new future for how we get around,” and people freaked out. And people thought, Google is going to solve this ASAP, and that maybe there’s an opportunity to beat them to the punch. It wasn’t that the venture capitalists saw a clear path for the technology to get from where it is to a self-driving car in every garage. But I think they thought that if they could get there, the rewards were so great that they were willing to bet on even the smallest chance of getting there. And I think all of these things came together. And everyone just got way ahead of themselves. And I think for some time now it’s been clear that a lot of that was unrealistic. But it’s very hard once these bubbles get inflated to walk them back. And so what has to happen is you have to have the bubbles pop and that’s what things like Argo AI going under was. That’s what that is, it’s a bubble pop.

Stephen Lacey  19:29

So, when a bubble pops, very often there’s direct financial pain. But there can be a lot of good that comes out of a bubble popping – you get rid of a lot of the froth and other ideas come in and people reset expectations, and what I’m hearing you say is that that will probably happen here. What good do you think will come from this as we do reset expectations and the industry sort of rallies around a new vision for autonomy or, or people understand where the technology actually is?

Ed Niedermeyer  20:01

One of the things that is unquestionably positive that’s come out of this bubble is that it has pushed the state of this technology forward. And that is something that happens; sometimes you have to have excessive spending in order to really push technology forward. In fact, usually that’s the case, right? Moonshots are not capital efficient, fundamentally. You’re sacrificing efficiency to push through something. And absolutely that is something that has happened over the last 10 years, the technology has taken massive steps forward. And that technology is now going towards the use cases where more economic value is going to be created in the short term. But that’s happening outside the car companies. When you look at the car companies, I think that’s where I get nervous, because I think they see in the short run their financial incentives are aligned behind selling people the perception of self-driving in their cars. I think that’s where Tesla has been successful is selling the idea that your car is going to drive itself. And the reason Tesla’s been successful at that is not because it’s technically plausible at all. It’s because it fits exactly what’s in the mental model of the American consumer, when they hear the word self-driving cars. And Tesla is the only company that’s selling them. And it’s a scam. But it’s been better business than anybody else in the driving automation tech sector. My concern is that the companies were now taking these technologies from these level four companies and putting them into privately on cars, they’re not going to make those cars drive themselves truly, right? At best, you get what sort of level three, which Ford and others have now been talking about, and Mercedes has a system like this out already. Level three essentially means it is self-driving, but only for certain stretches of road. So on certain mapped freeways for a certain number of miles in certain weather, the car will actually drive itself. Then of course, you also have what’s called the level two systems where the car looks like it’s driving itself but it actually isn’t because you need to be monitoring it actively the entire time. It is level two, level three human in-the-loop systems are where the auto industry is going to be for the for the near future. And there’s clearly a market for that, because self-driving is perceived as cool and leading edge and there’s a lot of cachet around being on the cutting edge of technology. The reality is, though, is that those systems don’t really deliver the value of self-driving technology of autonomous technology, because the human has to be in the loop. And you know, five years from now, 12 years from now, we may be sitting here talking about what a disaster the idea it was to make people think that the cars they buy and own are kind of driving themselves when they aren’t actually really. I think that’s going to create a lot of problems if we really go down that path all the way. And I think Tesla’s experience shows what some of those problems look like. And they’re ugly, because essentially, what you’re doing is you’re sort of encouraging people to believe something that’s not true, which and then endangers them, and sticks them with the responsibility for it, which I think is ethically and legally, really problematic. And if that keeps happening, what will happen is it will destroy people’s trust in this technology. So I think there’s plenty of drama for this technology ahead. I think there’s plenty of bad choices to be made and hard lessons to be learned. And I think that most of those cluster around where we try and fit this technology with cars as we know them. Where it goes from here is the hard part. There are companies like Amazon, for example, where Amazon’s business model is much better aligned with level-four autonomous vehicles, with robotic mobility infrastructure, especially in the supply chain logistics realm. I think there may well be other companies like that, who have not even really gotten into the space because they still think it’s a car thing. And they still don’t understand, “oh no, this is probably going to transform supply chain logistics, and countless other non-consumer facing applications before it ever changes cars.” But again, I think there’s a lot of individuals, people, companies, entities who still fundamentally misunderstand this technology because the car thing has just obfuscated its true nature so completely.

Stephen Lacey  24:13

Ed Niedermeyer is a journalist and author of Ludicrous, the Unvarnished Story of Tesla Motors, and the co-host of the Autonicast. Ed thanks so much for putting this into perspective for us.

Ed Niedermeyer  24:22

Absolutely. It was a pleasure speaking with you.

Stephen Lacey  24:27

That’s it for the show. The Carbon Copy is a co-production of Post Script Media and Canary Media. Thanks, everybody for listening. Can you hook us up with a rating and review on Apple or Spotify? We ask every time, I know it probably gets old, but it’s hugely helpful for us. And if you like what we’re doing here, share a link with a friend or colleague. You can go over to Post Script Media for transcripts of the show. You can also go to to get their newsletter and all the other great clean tech reporting that they’re doing there. And this episode was produced by me and In Alexandria Herr. Sean Marquand is our engineer. He also composed our theme song. Original music came from Echo Finch and Blue Dot Sessions. Post Script Media is supported by Prelude Ventures, a venture capital firm that partners with entrepreneurs to address climate change across a wide range of sectors: advanced energy, food and agriculture, transportation, logistics, advanced materials, manufacturing, and advanced computing. Thanks so much for being here.

The Carbon Copy is supported by FischTank PR, a public relations, strategic messaging, and social media agency dedicated to elevating the work of climate and clean energy companies. Learn more about FischTank’s approach to cleantech and their services:

The Carbon Copy is supported by Scale Microgrids, the distributed energy company dedicated to transforming the way modern energy infrastructure is designed, constructed, and financed. Distributed generation can be complex. Scale makes it easy. Learn more:

Learn more about your ad choices. Visit