Ford wants to have a hand in every part of the self-driving-car industry, including the software that replaces the human driver, the platform that connects and controls the cars and the services that spring up around them

Excerpt from 10 Nov 2017

…Ford wants to have a hand in every part of the self-driving-car industry, including the software that replaces the human driver, the platform that connects and controls the cars and the services that spring up around them. To that end, Ford has spent the past several years quietly snapping up tech talent. It struck a partnership with Lyft and acquired Chariot, a San Francisco-based start-up that runs group shuttles for commuters. It set up an office in Palo Alto that now has 205 employees and created Greenfield Labs, a business incubator. It invested in Argo AI, a Pittsburgh-based artificial-intelligence start-up, and several other companies, including Velodyne, which makes Lidar sensors, and a connected-car software outfit called Autonomic.

Most industry observers believe these are the right moves for Ford to be making. But it’s still tough to reconcile today’s Ford — which makes money by selling millions of combustion-engine trucks and S.U.V.s every year, along with a handful of sedans and hybrids — with the eco-tech-mobility conglomerate Hackett envisions it becoming. One lesson Hackett learned from his studies is that in order to stay competitive, businesses often have to give up the things that made them great in the first place. (He calls this the “perversity law.”) I asked him if this meant that Ford would need to stop manufacturing conventional, gas-powered vehicles to survive.

“Well, part of that statement might be right,” he said. “Right now, that doesn’t make sense.” He clarified, sort of: “I’m really just trying to say the better form of problem solving is to abstract the problems.”

So let’s try abstracting Ford’s problems, to see how they might be solved. First: Ford was born at the beginning of a now-outmoded era of industrial production, one that it arguably ushered in. Back then, companies like Ford simply manufactured goods, and customers bought them. Those transactions represented a change of ownership — a customer who bought a car from Ford was free to install 15-inch subwoofers in it, or paint it bright yellow, or use it in a demolition derby — and they were typically one-time sales. Ford’s responsibility for the car, and its opportunity to make a profit, ended the minute the warranty ran out.

The internet changed that, by allowing companies to wrap physical goods in digital services that can be sold again and again, at much higher margins. Instead of selling servers to corporate I.T. departments, Amazon sells them time shares on a centralized server farm; instead of buying pricey home solar panels, customers can lease them from a solar service, paying for only the power they use. Makers of thermostats, baby monitors and other household gadgets “extend the value chain” by charging customers a monthly fee for information about their home energy use or their babies’ sleep patterns. Even the language these new enterprises use is different. Goods-based businesses spoke in terms of revenues and profits. Service-based businesses count metrics like “churn,” “average revenue per user” and “lifetime value” — the total amount of money they can wring from a customer.

An Argo AI employee calibrating the sensors on a Ford Fusion test vehicle.Mark Peterson/Redux, for The New York Times

This phenomenon — which is sometimes referred to as “everything as a service,” abbreviated as XaaS and pronounced “zass,” among the kinds of people who say these things out loud — has quietly reshaped the entire economy. Trucking companies that once bought Michelin tires now rent them through a pay-per-mile service. Airlines that buy $15 million jet engines from General Electric can subscribe to OnPoint, a “power by the hour” program that includes maintenance costs and data-analysis tools. G.E. Aviation now makes the better part of its revenue from services like these.

The auto industry was slow to come to grips with this new model — in part because it long ago gave away the entire service piece of its business to gas stations, body shops, Jiffy Lubes and a hundred other barnacle industries. Thanks to old and byzantine state laws, most carmakers aren’t even allowed to sell or lease directly to their customers — every transaction has to go through an independent dealer. When it comes to data, automakers are flying blind. As Reilly P. Brennan, a venture capitalist who invests in self-driving technology, put it to me: “The urinal in the men’s bathroom knows more about its users than most carmakers in 2017.”

With self-driving cars, automakers see a chance to fix that. Ford’s executives talk hopefully about one day offering their own suite of software inside their autonomous vehicles, and building interfaces that companies like Netflix and Skype could plug into. They don’t see themselves usurping Silicon Valley, exactly, but they do see themselves playing in the same league. “Two to three years ago, the tech community was holding the auto industry at arm’s length,” said Ken Washington, Ford’s chief technology officer. “They thought we were sort of this curious industry that they could come into and disrupt.” He went on: “Now, when I go to Silicon Valley and I interact with tech companies, it’s a dynamic of much more intense interest in collaborating with us.”

Collaboration takes many forms, of course, and not all tech companies seem interested in an equal partnership. John Krafcik, the chief executive of Waymo, told me that while traditional auto companies “bring a lot to the party” in terms of hardware expertise, they’re not well equipped to solve the software-based problem of replacing a human driver, which requires huge teams of highly skilled engineers and tremendous amounts of technological infrastructure — the kinds of tools available to only a Google sibling company. “The self-driving aspect is really supercomplicated,” Krafcik said. “We’ve spent the last nine years trying to solve that puzzle. It requires a lot of skills.” It also requires focus, which is a problem for companies that still sell millions of conventional cars and trucks a year, that have thousands of employees to pay and quarterly results to deliver, that can’t exactly drop everything and become software companies.

In Detroit, cautious optimism prevails. But outside Detroit, there are real doubts that any auto company can transform itself quickly enough. “Autonomous vehicles are not an incremental shift,” said Carol Reiley, a founder of “It’s a very transformative shift, almost like moving from the horse and carriage to the car.” Like many tech executives, Reiley saw Detroit’s quest to control the self-driving-car industry as tactically sound but ultimately a long shot. “I don’t know if they can make it,” she said. “There’s a lot of baggage.”

An Argo AI team member completing final configurations on a self-driving test vehicle.Mark Peterson/Redux, for The New York Times

A week after my Detroit trip, I visited the Pittsburgh offices of Argo AI, the artificial-intelligence start-up that is creating the software brains for Ford’s self-driving vehicles. Ford is investing $1 billion for a majority stake and controls two of Argo’s five board seats. Which means that Argo is an independent company, but only barely — 40 of its 230 employees are transfers from Ford, and Ford is its only corporate partner.

Argo’s headquarters is in Pittsburgh’s Strip District, a formerly industrial area just outside downtown that has become a center of driverless-car activity. (Uber’s Advanced Technologies Group is down the street.) Its office is a high-ceilinged space in a modern office building with conference rooms named for famous scientists: Planck, Descartes, Hopper. When I arrived, engineers were eating catered lunch at their standing desks. Squint and ignore the smokestacks outside, and you could be in San Francisco — which is sort of the point. Ford wants Argo to look and feel like a nimble tech start-up, not a subsidiary of a century-old auto company.

I was greeted by Bryan Salesky and Peter Rander, Argo AI’s founders, both well-known figures in the small self-driving-car world. They had been special-faculty members at Carnegie Mellon’s prestigious robotics-engineering program, and Salesky competed in the 2007 Darpa Urban Challenge, a storied self-driving-vehicle race whose alumni now populate the top ranks of the autonomous-vehicle industry. Salesky went to Google to build self-driving cars; Rander went to Uber. They quit their jobs and started Argo AI in November of last year, and in February, with no real product to speak of, and hardly any employees, they announced that they had raised huge sums of money from Ford.

The day before my visit, Salesky made waves with a bombastic post on the company’s blog, in which he accused the autonomous-vehicle industry of overselling its progress. “Those who think fully self-driving vehicles will be ubiquitous on city streets months from now or even in a few years,” he wrote, “are not well connected to the state of the art or committed to the safe deployment of the technology.” It was a rare role reversal — the software engineer throwing up caution flags — and, perhaps, evidence that Ford is trying to stall for time. Officially, Argo is still aiming for Ford’s 2021 target, but Salesky waffled on the firmness of that deadline, saying there’s a chance a true Level 4 autonomous vehicle could be finished later, or possibly sooner. “It’s ready when it’s ready,” he told me.

Salesky freely admitted that he is playing catch-up. Ford won’t share statistics about its self-driving test cars, but public records from California give some sense of what it’s up against. Last year, Ford, which does most of its testing in Michigan and Arizona, logged just 590 autonomous miles in the state, while Waymo’s vehicles logged 635,867 miles. Argo is also building its code from scratch and doesn’t have access to the digital infrastructure that competitors like Waymo and Uber do. “We’re not out there trying to tell everyone we’re going to be first or beat X, Y or Z company,” Salesky said. “The goal is to get to market with something that can scale up.”

Driving isn’t just a mechanical task — it’s a social act, and in order to coexist with human drivers, self-driving cars will need to develop a level of social awareness.  But, he reiterated, this is a hard problem. Driving is a miracle of neurobiological coordination that involves the flawless simultaneous execution of a billion tiny and subtle reflexes, and it turns out that’s a more complicated task for a computer than teaching it to play chess or compete on game shows. A self-driving car has to correctly identify and label millions of objects, understand city layouts and traffic laws and operate in a variety of road conditions. It has to be taught to handle everyday driving hazards (high-speed merges) and rarer incidents (objects in the road), as well as issues that would never affect a human driver (a chunk of debris that flies up and knocks out a sensor).

In order to work properly, a self-driving car also has to understand how humans behave. It needs to know the difference between a car that is idling in the right-hand lane (in which case the autonomous vehicle should steer around it) and one that is about to parallel park (in which case the vehicle should stay in its lane, giving the other car some room). It needs to predict that the jogger running toward the corner will stop for traffic, but that the kid running to chase a basketball might not. It needs to be able to navigate a four-way stop, which in polite parts of the country involves lots of eye contact and you-first hand gestures. “This goes beyond just seeing and understanding the world,” Salesky said. “It means understanding what each of the actors in the world is going to do.”

In other words, driving isn’t just a mechanical task — it’s a social act, and in order to coexist with human drivers, self-driving cars will need to develop a level of social awareness that approaches that of a full-fledged A.I. This is why some computer scientists think Waymo is going to be unbeatable. As an Alphabet company, Waymo has access to some of the best engineering talent in A.I., along with Google’s enormous data troves and the machine-learning systems that have been trained using that data. In other words, the vast data-harvesting powers that make Google’s consumer products so creepy could also make Waymo’s vehicles better at navigating the physical world.

But all this obsessive focus on tech elides a much more foundational question about our relationship to cars. Tech companies tend to view driving as a functional problem — get safely from Point A to Point B as quickly as possible — in which the vehicle itself is largely an extraneous variable. A car might begin its life as a Toyota or a Hyundai, but once it’s autonomous and on a ride-sharing network, it’s just an Uber. Silicon Valley believes that software, not hardware, is what matters to riders, and this will be especially true of self-driving cars — which, at least in early days, will most likely be shared among strangers. 

But Detroit has spent decades and billions of advertising dollars persuading us that its mass-produced cars are anything but homogeneous metal boxes — that they carry profound talismanic qualities that transfer to their owners. You can be an F-150 woman or a Mustang man, and those choices signal tribal affiliations that extend far past the act of driving. Ford’s challenge, then, is keeping these storied brands alive for the consumers who still obsess over them, while Argo builds the software that will satisfy the car-agnostic city dwellers. It’s a tricky balancing act, made even more complicated by the politics of a 114-year-old Detroit manufacturing company lashing its fate to a one-year-old Pittsburgh tech start-up, and vice versa. “They’ve put a lot of trust in us,” Salesky said. “We’ve also put a lot of trust in them.”

Argo AI’s self-driving test vehicles at its facility in Dearborn, Mich. Mark Peterson/Redux, for The New York Times

THE HALL CONNECTING the C-suites at Ford’s headquarters was a mess when I visited in October: exposed wiring, plastic sheeting hanging from the walls. As I made my way to Bill Ford’s office, my P.R. escorts offered an explanation. Parts of the floor were being reimagined as an open-plan creative space, and the chaos had spilled outward. When we reached the end of the hall, Ford welcomed me into his office, a midcentury wooden cavern with a fish tank built into one wall and a panoramic view of Dearborn along the other. The dreary vista stood in stark relief with the golden-era artifacts in Ford’s office: the replica of a 1956-57 Continental Mark II that was designed by his father, William Clay Ford Sr.; a desk that belonged to his grandfather, Edsel Ford.

Bill Ford has been working in the family business for 38 years and spent most of that time as an iconoclast. His interest in “smart mobility” — which amounts to an interest in reducing the number of cars on the road — was until recently considered heretical. But the past few years have vindicated his predictions. He thinks that the coming convergence of tech and transportation could turn Detroit and Silicon Valley into permanent collaborators. He started Fontinalis Partners, a venture-capital fund outside Ford, to invest in mobility start-ups, and flies regularly to San Francisco to meet with entrepreneurs working on transportation-related projects.

“The narrative, probably even two years ago, would have been ‘tech companies win, autos lose,’ ” he said. “What’s happened, though, is there’s been a realization that we have a lot more intellectual property than the world realized,” by which he means celebrated lines like the F-150, along with some proprietary under-the-hood technology that I won’t pretend to understand. “Second, we really know how to integrate a lot of what the tech companies have into our vehicles, and we know how to make vehicles as well.”

I was struck by how frequently this simple point is deployed at Ford: We know how to make stuff, and making stuff isn’t easy. At first, I dismissed it as a naïve attempt to steer a conversation about Detroit’s shortcomings back toward its strengths. Apple taught suppliers halfway across the world how to make iPhones with micron-level precision — how hard could it be to slap some metal together and put it on wheels?

But making cars isn’t like making other things. It’s a dizzying logistical ballet that involves heavy machinery, grueling labor and countless opportunities for error. It’s wrangling union contracts while keeping the line moving. It’s paying people enough to ensure that they won’t miss a loose bolt and kill someone. It’s designing objects that not only work well but also fill a psychological need for buyers. In the end, maybe the self-driving-car race will be decided not purely on technological achievement, but also on these much more prosaic questions — who can make good things and persuade people to use them? After all, the Model T wasn’t the first car invented, or the best at the time. But it was accessible and functional and spoke to people’s needs in a simple way during a time of confusing technological change. That conversation is about to begin again.