Jon McNeill — a former Tesla and Lyft executive and current CEO at DVx Ventures — spread the word last week at the World Business Forum on what he calls the “Tesla innovation method” of radical simplification. 

After his presentation, we sat down to chat about Tesla’s automation goals, its new robotaxi, and why building electric vehicles is a tough business. 

On the Tesla robotaxi

Tesla rolled out Autopilot, its advanced driver assistance system (ADAS) that can perform limited automated driving tasks, back in 2015. McNeill said automation has been a major initiative for Tesla for almost a decade. 

“It was a topic of conversation every week, if not every day,” he told TechCrunch.

That’s partly why McNeill, along with many others, was surprised by the nothing burger that was Tesla’s much-anticipated robotaxi reveal earlier this month. 

“Oftentimes, Elon has got something up his sleeve where he’ll not reveal it until one of those product events. And so I was really wondering, what’s up his sleeve? What does he have? Because Waymo and Cruise are already at Level 4. Tesla’s at Level 2. He’s got to have some major breakthroughs he’s going to talk about to take it from Level 2 to Level 4.”

(The SAE defines Level 2 automation as ADAS that can do things like brake support and lane centering, but require a human to constantly drive and supervise. Level 4 means the car can drive itself without a human under certain conditions.)

But in the end, Musk only revealed 20 robotaxi prototypes. “That’s not to say they don’t have something up their sleeve still, but I was a little surprised that we just got a prototype.”

McNeill said he was expecting Tesla to reveal a roadmap, to explain what will happen and when in order to really build confidence in the brand. And confidence, he says, comes through more specifics. 

(After our interview during Tesla’s third-quarter earnings call, Musk said he hopes to launch an autonomous ride-hail service in California and Texas in 2025.)

On automation through vision only

McNeill’s presentation was all about how Tesla has been successful through simplifying processes. I asked him if Tesla’s choice to pursue self-driving through only cameras, rather than relying on lidar and radar, was a sort of false simplification. Yes, using only cameras might simplify the sensor stack, but it creates a lot more work on the back end. 

On top of being an automaker, an EV charging company, and a solar company, Tesla is now building Dojo, a supercomputer that it hopes will help it train the neural nets necessary to achieve full self-driving through vision alone. And Dojo is a bet that Musk has admitted might not pan out. 

McNeill expressed his concern for Tesla’s vision-only approach. “Human beings can drive a car, and we only have two eyes. So if you give a car eight eyes, it ought to be able to solve the problem, right? But I think that misses part of the challenge, and that is, there are things that cause us not to be able to see,” he said. Like snow, fog, sun glare, darkness, and other objects. 

“Having eyes may not be sufficient. Lidar can see through all those things … and I think when people’s lives are at stake, you don’t cut costs or pinch pennies because there’s a safety issue. And so some people say you can solve it just with vision, but I wonder if that’s really true because I’m not sure human vision is adequate. That’s why we have 40,000 traffic deaths a year.” 

Caveat: McNeill is on the board of GM’s Cruise, which commercialized Level 4 autonomy using lidar before its fleet was suspended last year following a safety incident. 

“If the eye can’t see through the dark, through the rain, around the corner, etc., I don’t know how supercompute solves that.”

On the idea of Tesla adopting lidar

Tesla has backed its vision-only approach for years. It hasn’t panned out yet, and it might not ever get there. I asked McNeill whether it would be crazy for Tesla to reintroduce lidar back into the system. Or has the automaker sunk its costs into cameras?

McNeill said lidar costs have come down so much that from a financial standpoint, it wouldn’t be a huge problem for Tesla to adopt the sensor. But from a reputational standpoint? McNeill said Tesla has a track record of not being fearful of the sunk cost fallacy or even frustrating the customer. 

He pointed to Musk’s statements from at least 2019 that any Tesla purchased then would have the compute and sensor set necessary for full self-driving when the feature became available. Tesla has had to walk back on that promise, and during the automaker’s third-quarter earnings call, Musk said vehicles with outdated hardware will receive a free retrofit to accommodate unsupervised FSD. 

On why EV startups fail today

Fisker is going through a messy bankruptcy. Rivian is losing roughly $30,000 per car it produces. Canoo has changed its business model so many times, we’ve lost track. In short, building cars from the ground up is hard. Do you need someone as wild as Elon Musk at the helm to be successful? I asked. 

“I do think you need a crazy, effective leader to make it through the gauntlet of everything that you have to prove as a car company,” McNeill said. “It’s hard to describe how difficult it is to take 10,000 parts from 72 countries and make a car out of it, and then put the software together and make the car run. Like manufacturing these things is really, really, really the hard part.”

Reflecting on his disappointment with Tesla’s robotaxi event, McNeill said that it’s really easy to produce a prototype, but it’s really hard to produce at mass scale. 

“That’s where the car industry is incredibly brutal, and even for incumbents,” he said. “You see Stellantis now really struggling. You see Volkswagen really struggling. It’s because the fixed costs are so high. If you don’t get the product right and you don’t sell it, it’s merciless.”