Note: While reading a book whenever I come across something interesting, I highlight it on my Kindle. Later I turn those highlights into a blogpost. It is not a complete summary of the book. These are my notes which I intend to go back to later. Let’s start!

  • When Kimbal moved to Canada and joined Elon as a student at Queen’s, the brothers developed a routine. They would read the newspaper and pick out the person they found most interesting. Elon was not one of those eager-beaver types who liked to attract and charm mentors, so the more gregarious Kimbal took the lead in cold-calling the person. “If we were able to get through on the phone, they usually would have lunch with us,” he says.

  • Nicholson says that Scotiabank was navigating the Latin American debt situation using its own methods, which worked better. “He came away with an impression that the bank was a lot dumber than in fact it was,” Nicholson says. “But that was a good thing, because it gave him a healthy disrespect for the financial industry and the audacity to eventually start what became PayPal.” Musk also drew another lesson from his time at Scotiabank: he did not like, nor was he good at, working for other people. It was not in his nature to be deferential or to assume that others might know more than he did.

  • When the other engineers went home, Musk would sometimes take the code they were working on and rewrite it. With his weak empathy gene, he didn’t realize or care that correcting someone publicly—or, as he put it, “fixing their fucking stupid code”—was not a path to endearment. He had never been a captain of a sports team or the leader of a gang of friends, and he lacked an instinct for camaraderie. Like Steve Jobs, he genuinely did not care if he offended or intimidated the people he worked with, as long as he drove them to accomplish feats they thought were impossible. “It’s not your job to make people on your team love you,” he said at a SpaceX executive session years later. “In fact, that’s counterproductive.” He was toughest on Kimbal. “I love, love, love my brother very much, but working with him was hard,” Kimbal says. Their disagreements often led to rolling-on-the-office-floor fights. They fought over major strategy, minor slights, and the name Zip2. (Kimbal and a marketing firm came up with it; Elon hated it.) “Growing up in South Africa, fighting was normal,” Elon says. “It was part of the culture.” They had no private offices, just cubicles, so everyone had to watch. In one of their worst fights, they wrestled to the floor and Elon seemed ready to punch Kimbal in the face, so Kimbal bit his hand and tore off a hunk of flesh. Elon had to go to the emergency room for stitches and a tetanus shot. “When we had intense stress, we just didn’t notice anyone else around us,” says Kimbal. He later admitted that Elon was right about Zip2. “It was a shitty name.”

  • One of Musk’s management tactics, then as later, was to set an insane deadline and drive colleagues to meet it. He did that in the fall of 1999 by announcing, in what one engineer called “a dick move,” that would launch to the public on Thanksgiving weekend. In the weeks leading up to that, Musk prowled the office each day, including Thanksgiving, in a nervous and nervous-making frenzy, and slept under his desk most nights. One of the engineers who went home at 2 a.m. Thanksgiving morning got a call from Musk at 11 a.m. asking him to come back in because another engineer had worked all night and was “not running on full thrusters anymore.” Such behavior produced drama and resentments, but also success. When the product went live that weekend, all the employees marched to a nearby ATM, where Musk inserted an debit card. Cash whirred out and the team celebrated.

  • That made Musk fundamentally different from Thiel, who always focused on limiting risks. He and Hoffman once planned to write a book on their experience at PayPal. The chapter on Musk was going to be titled “The Man Who Didn’t Understand the Meaning of the Word ‘Risk.’ ” Risk addiction can be useful when it comes to driving people to do what seems impossible. “He’s amazingly successful getting people to march across a desert,” Hoffman says. “He has a level of certainty that causes him to put all of his chips on the table.” That was more than just a metaphor. Many years later, Levchin was at a friend’s bachelor pad hanging out with Musk. Some people were playing a high-stakes game of Texas Hold ’Em. Although Musk was not a card player, he pulled up to the table. “There were all these nerds and sharpsters who were good at memorizing cards and calculating odds,” Levchin says. “Elon just proceeded to go all in on every hand and lose. Then he would buy more chips and double down. Eventually, after losing many hands, he went all in and won. Then he said, ‘Right, fine, I’m done.’ ” It would be a theme in his life: avoid taking chips off the table; keep risking them. That would turn out to be a good strategy. “Look at the two companies he went on to build, SpaceX and Tesla,” says Thiel. “Silicon Valley wisdom would be that these were both incredibly crazy bets. But if two crazy companies work that everyone thought couldn’t possibly work, then you say to yourself, ‘I think Elon understands something about risk that everybody else doesn’t.’ ”

  • As he stewed about the absurd price the Russians wanted to charge, he employed some first-principles thinking, drilling down to the basic physics of the situation and building up from there. This led him to develop what he called an “idiot index,” which calculated how much more costly a finished product was than the cost of its basic materials. If a product had a high idiot index, its cost could be reduced significantly by devising more efficient manufacturing techniques. Rockets had an extremely high idiot index. Musk began calculating the cost of carbon fiber, metal, fuel, and other materials that went into them. The finished product, using the current manufacturing methods, cost at least fifty times more than that. If humanity was going to get to Mars, the technology of rockets must radically improve. And relying on used rockets, especially old ones from Russia, was not going to push the technology forward. So on the flight home, he pulled out his computer and started making spreadsheets that detailed all of the materials and costs for building a midsize rocket. Cantrell and Griffin, sitting in the row behind him, ordered drinks and laughed. “What the fuck do you think that idiot-savant is doing up there?” Griffin asked Cantrell. Musk turned around and gave them an answer. “Hey, guys,” he said, showing them the spreadsheet, “I think we can build this rocket ourselves.” When Cantrell looked at the numbers, he said to himself, “I’ll be damned—that’s why he’s been borrowing all my books.” Then he asked the flight attendant for another drink.

  • One thing that Mueller insisted on was that Musk put two years’ worth of compensation into escrow. He was not an internet millionaire, and he did not want to take the chance of being unpaid if the venture failed. Musk agreed. It did, however, cause him to consider Mueller an employee rather than a cofounder of SpaceX. It was a fight he had regarding PayPal and would have again involving Tesla. If you’re unwilling to invest in a company, he felt, you shouldn’t qualify as a founder. “You cannot ask for two years of salary in escrow and consider yourself a cofounder,” he says. “There’s got to be some combination of inspiration, perspiration, and risk to be a cofounder.”

  • As his team grew, Musk infused it with his tolerance for risk and reality-bending willfulness. “If you were negative or thought something couldn’t be done, you were not invited to the next meeting,” Mueller recalls. “He just wanted people who would make things happen.” It was a good way to drive people to do what they thought was impossible. But it was also a good way to become surrounded by people afraid to give you bad news or question a decision.

  • Musk was laser-focused on keeping down costs. It was not simply because his own money was on the line, though that was a factor. It was also because cost-effectiveness was critical for his ultimate goal, which was to colonize Mars. He challenged the prices that aerospace suppliers charged for components, which were usually ten times higher than similar parts in the auto industry. His focus on cost, as well as his natural controlling instincts, led him to want to manufacture as many components as possible in-house, rather than buy them from suppliers, which was then the standard practice in the rocket and car industries. At one point SpaceX needed a valve, Mueller recalls, and the supplier said it would cost $250,000. Musk declared that insane and told Mueller they should make it themselves. They were able to do so in months at a fraction of the cost. Another supplier quoted a price of $120,000 for an actuator that would swivel the nozzle of the upper-stage engines. Musk declared it was not more complicated than a garage door opener, and he told one of his engineers to make it for $5,000. Jeremy Hollman, one of the young engineers working for Mueller, discovered that a valve that was used to mix liquids in a car wash system could be modified to work with rocket fuel.

  • When SpaceX began producing its first Merlin engines, Musk asked Mueller how much they weighed. About a thousand pounds, Mueller responded. The Tesla Model S engine, Musk said, weighed about four thousand pounds and cost about $30,000 to make. “So if the Tesla engine is four times as heavy as your engine, why does yours cost so fucking much?” One reason was that rocket components were subject to hundreds of specifications and requirements mandated by the military and NASA. At big aerospace companies, engineers followed these religiously. Musk did the opposite: he made his engineers question all specifications. This would later become step one in a five-point checklist, dubbed “the algorithm,” that became his oft-repeated mantra when developing products. Whenever one of his engineers cited “a requirement” as a reason for doing something, Musk would grill them: Who made that requirement? And answering “The military” or “The legal department” was not good enough. Musk would insist that they know the name of the actual person who made the requirement. “We would talk about how we were going to qualify an engine or certify a fuel tank, and he would ask, ‘Why do we have to do that?’ ” says Tim Buzza, a refugee from Boeing who would become SpaceX’s vice president of launch and testing. “And we would say, ‘There is a military specification that says it’s a requirement.’ And he’d reply, ‘Who wrote that? Why does it make sense?’ ” All requirements should be treated as recommendations, he repeatedly instructed. The only immutable ones were those decreed by the laws of physics.

  • When Mueller was working on the Merlin engines, he presented an aggressive schedule for completing one of the versions. It wasn’t aggressive enough for Musk. “How the fuck can it take so long?” he asked. “This is stupid. Cut it in half.” Mueller balked. “You can’t just take a schedule that we already cut in half and then cut it in half again,” he said. Musk looked at him coldly and told him to stay behind after the meeting. When they were alone, he asked Mueller whether he wanted to remain in charge of engines. When Mueller said he did, Musk replied, “Then when I ask for something, you fucking give it to me.” Mueller agreed and arbitrarily cut the schedule in half. “And guess what?” he says. “We ended up developing it in about the time that we had put in that original schedule.” Sometimes Musk’s insane schedules produced the impossible, sometimes they didn’t. “I learned never to tell him no,” Mueller says. “Just say you’re going to try, then later explain why if it doesn’t work out.” Musk insisted on setting unrealistic deadlines even when they weren’t necessary, such as when he ordered test stands to be erected in weeks for rocket engines that had not yet been built. “A maniacal sense of urgency is our operating principle,” he repeatedly declared. The sense of urgency was good for its own sake. It made his engineers engage in first-principles thinking. But as Mueller points out, it was also corrosive. “If you set an aggressive schedule that people think they might be able to make, they will try to put out extra effort,” he says. “But if you give them a schedule that’s physically impossible, engineers aren’t stupid. You’ve demoralized them. It’s Elon’s biggest weakness.” Steve Jobs did something similar. His colleagues called it his reality-distortion field. He set unrealistic deadlines, and when people balked, he would stare at them without blinking and say, “Don’t be afraid, you can do it.” Although the practice demoralized people, they ended up accomplishing things that other companies couldn’t. “Even though we failed to meet most schedules or cost targets that Elon laid out, we still beat all of our peers,” Mueller admits. “We developed the lowest-cost, most awesome rockets in history, and we would end up feeling pretty good about it, even if Dad wasn’t always happy with us.”

  • Musk took an iterative approach to design. Rockets and engines would be quickly prototyped, tested, blown up, revised, and tried again, until finally something worked. Move fast, blow things up, repeat. “It’s not how well you avoid problems,” Mueller says. “It’s how fast you figure out what the problem is and fix it.” For example, there was a set of military specifications on how many hours each new version of an engine needed to be test-fired under a long list of different conditions. “It was a tedious approach and very expensive,” Tim Buzza explains. “Elon told us just to build one engine and fire it up on the test stand; if it worked, put it on a rocket and fly it.” Because SpaceX was a private company, and because Musk was willing to flout rules, it could take the risks it wanted. Buzza and Mueller pushed their engines until they broke, and then said, “Okay, now we know what the limits are.”

  • Musk has a rule about responsibility: every part, every process, and every specification needs to have a name attached. He can be quick to personalize blame when something goes wrong. In the case of the launch failure, it became evident that the leak had come from a small B-nut that secured a fuel line. Musk fingered an engineer named Jeremy Hollman, one of Mueller’s first hires, who, the night before the launch, had removed and then reattached the nut in order to get access to a valve. At a public symposium a few days later, Musk described the mistake by “one of our most experienced technicians,” and insiders knew he was referring to Hollman. Hollman had stayed behind in Kwaj for two weeks to analyze the debris. On his flight from Honolulu to Los Angeles, he was reading news stories about the failure and was shocked to see that Musk had blamed him. As soon as he landed, he drove the two miles from the airport to SpaceX headquarters and barged into Musk’s cubicle. A shouting match erupted, and both Shotwell and Mueller went over to try to calm things down. Hollman wanted the company to retract Musk’s statement, and Mueller pressed for permission to do so. “I’m the CEO,” Musk replied. “I’m the one that deals with the press, so stay out of it.” Hollman told Mueller he would stay at the company only if he never had to deal directly with Musk. He left SpaceX a year later. Musk says he doesn’t remember the event, but he adds that Hollman was not a great engineer. Mueller disagrees: “We lost a good guy.” As it turned out, Hollman was not at fault. When the fuel line was found, part of the B-nut was still attached, but it was corroded and had cracked in half. The sea air of Kwaj was to blame. Because Gracias spoke Spanish like most of the factory workers, he was able to learn from them where the problems were. “I realized that if you invest in a company, you should spend all your time on the shop floor,” he says. When he asked how they could speed things up, one of the workers explained that having smaller vats for the nickel baths would make the plating go faster. Those and other worker-generated ideas succeeded so well that the factory began turning a profit, and Gracias started buying more troubled companies. He learned one very big lesson from these ventures: “It’s not the product that leads to success. It’s the ability to make the product efficiently. It’s about building the machine that builds the machine. In other words, how do you design the factory?” It was a guiding principle that Musk would make his own.

  • Carl Hoffman, a Wired reporter who had watched the failure of the second launch with Musk, reached him to ask how he maintained his optimism. “Optimism, pessimism, fuck that,” Musk answered. “We’re going to make it happen. As God is my bloody witness, I’m hell-bent on making it work.”

  • Musk’s decision to reverse his orders about quality controls taught Buzza two things: Musk could pivot when situations changed, and he was willing to take more risk that anyone. “This is something that we had to learn, which was that Elon would make a statement, but then time would go on and he would realize, ‘Oh no, actually we can do it this other way,’ ” Buzza says.

  • Regularly prodded by Musk, Mosdell rebuilt the area in SpaceX’s typical scrappy way, literally. He and his boss, Tim Buzza, scavenged for components that could be cheaply repurposed. Buzza was driving down a road at Cape Canaveral and saw an old liquid oxygen tank. “I asked the general if we could buy it,” he says, “and we got a $1.5 million pressure vessel for scrap. It’s still at Pad 40.” Musk also saved money by questioning requirements. When he asked his team why it would cost $2 million to build a pair of cranes to lift the Falcon 9, he was shown all the safety regulations imposed by the Air Force. Most were obsolete, and Mosdell was able to convince the military to revise them. The cranes ended up costing $300,000. Decades of cost-plus contracts had made aerospace flabby. A valve in a rocket would cost thirty times more than a similar valve in a car, so Musk constantly pressed his team to source components from non-aerospace companies. The latches used by NASA in the Space Station cost $1,500 each. A SpaceX engineer was able to modify a latch used in a bathroom stall and create a locking mechanism that cost $30. When an engineer came to Musk’s cubicle and told him that the air-cooling system for the payload bay of the Falcon 9 would cost more than $3 million, he shouted over to Gwynne Shotwell in her adjacent cubicle to ask what an air-conditioning system for a house cost. About $6,000, she said. So the SpaceX team bought some commercial air-conditioning units and modified their pumps so they could work atop the rocket. When Mosdell worked for Lockheed and Boeing, he rebuilt a launchpad complex at the Cape for the Delta IV rocket. The similar one he built for the Falcon 9 cost one-tenth as much. SpaceX was not only privatizing space; it was upending its cost structure.

  • So Musk and Straubel decided to pursue a partnership with their battery supplier, Panasonic. Together they would build a facility where Panasonic would make the battery cells and then Tesla would turn them into battery packs for cars. The 10-million-square-foot factory would cost $5 billion, and Panasonic would finance $2 billion of it. But Panasonic’s top leaders were hesitant. They had never had that type of partnership, and Musk (understandably) did not strike them as an easy guy to dance with. To prod Panasonic, Musk and Straubel came up with a charade. At a site near Reno, Nevada, they set up lights and sent in bulldozers to start preparing for construction. Then Straubel invited his counterpart at Panasonic to join him on a viewing platform to watch the work. The message was clear: Tesla was forging ahead with the factory. Did Panasonic want to be left behind? It worked. Musk and Straubel were invited to Japan by Panasonic’s new young president Kazuhiro Tsuga. “It was a come-to-Jesus session where we had to make him truly commit that we were going to build the insane Gigafactory together,” Straubel says. The dinner was a formal, multicourse affair at a traditional low-table Japanese restaurant. Straubel was fearful about how Musk would behave. “Elon can be so much hell and brimstone in meetings and just unpredictable as all get out,” he says. “But I’ve also seen him flip a switch and suddenly be this incredibly effective, charismatic, high-emotional-intelligence business person, when he has to do it.” At the Panasonic dinner, the charming Musk appeared. He sketched out his vision for moving the world to electric vehicles and why the two companies should do it together. “I was mildly shocked and impressed, because, whoa, this is not like how Elon usually was on other days,” says Straubel. “He’s a person who’s all over the map, and you don’t know what he’s going to say or do. And then, all of a sudden, he pulls it all together.” At the dinner, Tsuga agreed to be a 40 percent partner in the Gigafactory. When asked why Panasonic decided to do the deal, he replied, “We are too conservative. We are a ninety-five-year-old company. We have to change. We have to use some of Elon’s thinking.”

  • “Elon and I hit it off right away, and I went to visit him at his rocket factory,” Hassabis says. While sitting in the canteen overlooking the assembly lines, Musk explained that his reason for building rockets that could go to Mars was that it might be a way to preserve human consciousness in the event of a world war, asteroid strike, or civilization collapse. Hassabis added another potential threat to the list: artificial intelligence. Machines could become superintelligent and surpass us mere mortals, perhaps even decide to dispose of us. Musk paused silently for almost a minute as he processed this possibility. During such trancelike periods, he says, he runs visual simulations about the ways that multiple factors may play out over the years. He decided that Hassabis might be right about the danger of AI, and he invested $5 million in DeepMind as a way to monitor what it was doing.

  • Musk flipped from being an apostle of automation to a new mission he pursued with similar zeal: find any part of the line where there was a holdup and see if de-automation would make it go faster. “We began sawing robots out of the production line and throwing them into the parking lot,” Straubel says. On one weekend, they marched through the factory painting marks on machinery to be jettisoned. “We put a hole in the side of the building just to remove all that equipment,” Musk says. The experience became a lesson that would become part of Musk’s production algorithm. Always wait until the end of designing a process—after you have questioned all the requirements and deleted unnecessary parts—before you introduce automation.

  • Musk had come to realize that designing a good factory was like designing a microchip. It was important to create, in each patch, the right density, flow, and processes. So he paid the most attention to a monitor that showed each station on the assembly line with a green or red light indicating whether it was flowing properly. There were also green and red lights at the stations themselves, so Musk was able to walk the floor and home in on trouble spots. His team called it “walk to the red.” The surge at Fremont began the first week of April 2018. That Monday, he began walking the floor with his fast bearlike gait, heading to any red light he saw. What’s the problem? A part was missing. Who’s in charge of that part? Get him over here. A sensor keeps tripping. Who calibrated it? Find someone who can open the console. Can we adjust the settings? Why do we even need that fucking sensor?

  • At another of the stations, the partially completed auto bodies were bolted to a skid that moved them through the final assembly process. The robotic arms tightening the bolts were, Musk thought, moving too slowly. “Even I could do it faster,” he said. He told the workers to see what the settings were for the bolt drivers. But nobody knew how to open the control console. “Okay,” he said, “I’m just going to just stand here until we find someone who can bring up that console.” Finally a technician was found who knew how to access the robot’s controls. Musk discovered that the robot was set to 20 percent of its maximum speed and that the default settings instructed the arm to turn the bolt backward twice before spinning it forward to tighten. “Factory settings are always idiotic,” he said. So he quickly rewrote the code to delete the backward turns. Then he set the speed to 100 percent capacity. That started to strip the threads, so he dialed it back to 70 percent. It worked fine and cut the time it took to bolt the cars to the skids by more than half.

  • The production lines often halted when safety sensors were triggered. Musk decided they were too sensitive, tripping when there was no real problem. He tested some of them to see if something small like a piece of paper falling past the sensor could trigger a stoppage. This led to a crusade to weed out sensors in both Tesla cars and SpaceX rockets. “Unless a sensor is absolutely needed to start an engine or safely stop an engine before it explodes, it must be deleted,” he wrote in an email to SpaceX engineers. “Going forward, anyone who puts a sensor (or anything) on the engine that isn’t obviously critical will be asked to leave.”

  • At any given production meeting, whether at Tesla or SpaceX, there is a nontrivial chance that Musk will intone, like a mantra, what he calls “the algorithm.” It was shaped by the lessons he learned during the production hell surges at the Nevada and Fremont factories. His executives sometimes move their lips and mouth the words, like they would chant the liturgy along with their priest. “I became a broken record on the algorithm,” Musk says. “But I think it’s helpful to say it to an annoying degree.” It had five commandments:

  1. Question every requirement. Each should come with the name of the person who made it. You should never accept that a requirement came from a department, such as from “the legal department” or “the safety department.” You need to know the name of the real person who made that requirement. Then you should question it, no matter how smart that person is. Requirements from smart people are the most dangerous, because people are less likely to question them. Always do so, even if the requirement came from me. Then make the requirements less dumb.
  2. Delete any part or process you can. You may have to add them back later. In fact, if you do not end up adding back at least 10% of them, then you didn’t delete enough.
  3. Simplify and optimize. This should come after step two. A common mistake is to simplify and optimize a part or a process that should not exist.
  4. Accelerate cycle time. Every process can be speeded up. But only do this after you have followed the first three steps. In the Tesla factory, I mistakenly spent a lot of time accelerating processes that I later realized should have been deleted.
  5. Automate. That comes last. The big mistake in Nevada and at Fremont was that I began by trying to automate every step. We should have waited until all the requirements had been questioned, parts and processes deleted, and the bugs were shaken out.
  • The algorithm was sometimes accompanied by a few corollaries, among them:

All technical managers must have hands-on experience. For example, managers of software teams must spend at least 20% of their time coding. Solar roof managers must spend time on the roofs doing installations. Otherwise, they are like a cavalry leader who can’t ride a horse or a general who can’t use a sword. Comradery is dangerous. It makes it hard for people to challenge each other’s work. There is a tendency to not want to throw a colleague under the bus. That needs to be avoided. It’s OK to be wrong. Just don’t be confident and wrong. Never ask your troops to do something you’re not willing to do. Whenever there are problems to solve, don’t just meet with your managers. Do a skip level, where you meet with the level right below your managers. When hiring, look for people with the right attitude. Skills can be taught. Attitude changes require a brain transplant. A maniacal sense of urgency is our operating principle. The only rules are the ones dictated by the laws of physics. Everything else is a recommendation.

  • Musk is usually not sentimental about people leaving. He likes fresh blood. He is more concerned with a phenomenon he calls “phoning in rich,” meaning people who have worked at the company for a long time and, because they have enough money and vacation homes, no longer hunger to stay all night on the factory floor. But in the case of Straubel, Musk felt a personal affection as well as professional trust. “I was a little surprised at Elon’s reluctance to have me leave,” says Straubel.

  • By July 2019, von Holzhausen and Morris had built a full-size mock-up of a futuristic, jarring, cyber design with sharp angles and diamond facets. One Friday they surprised Musk, who had not yet seen it, by putting it in the middle of the showroom floor next to the more traditional model they had been considering. When Musk walked in the door leading from the SpaceX factory, his reaction was instantaneous. “That’s it!” he exclaimed. “I love it. We are doing that. Yes, this is what we are going to do! Yes, okay, done.” It became known as the Cybertruck. “A majority of people in this studio hated it,” says von Holzhausen. “They were like, ‘You can’t be serious.’ They didn’t want to have anything to do with it. It was just too weird.” Some of the engineers started working secretly on an alternative version. Von Holzhausen, who is as gentle as Musk is brusque, spent time listening carefully to their concerns. “If you don’t have buy-in from the people around you, it’s hard to get things done,” he says. Musk was less patient. When some designers pushed him to at least do some market testing, Musk replied, “I don’t do focus groups.”

  • The design of the truck was finished in August 2019, Musk told the team he wanted to reveal a working prototype publicly that November—in three months rather than the nine months it normally takes to make a running prototype. “We won’t be able to have one that we can actually drive by then,” von Holzhausen said. Musk replied, “Yes, we will.” His unrealistic deadlines usually do not pan out, but in some cases they do. “It forced the team to come together, work twenty-four-seven, and rally around that date,” von Holzhausen says. On November 21, 2019, the truck was driven onto a stage in the design studio for a presentation to the press and invited guests. There were gasps. “Many in the crowd clearly couldn’t believe that this was actually the vehicle they’d come to see,” CNN reported. “The Cybertruck looks like a large metal trapezoid on wheels, more like an art piece than a truck.” There was also an unexpected surprise when von Holzhausen tried to show the toughness of the truck. He swung at the body with a sledgehammer, which didn’t make a dent. Then he threw a metal ball at one of the “armor glass” windows, to show it wouldn’t break. To his surprise, it cracked. “Oh my fucking God!” Musk said. “Well, maybe that was a little too hard.” Overall, the presentation was not a great success. Tesla stock dropped 6 percent the next day. But Musk was satisfied. “Trucks have been the same for a very long time, like for a hundred years,” he told the crowd. “We want to try something different.” Afterward, he took Grimes for a spin in the prototype to Nobu restaurant, where the valet parkers just stared at it without touching. On the way out, pursued by paparazzi, he drove over a pylon in the parking lot with a “No Left Turn” sign and turned left. In the midst of the hellacious summer of 2018, Musk was having a Spidey sense that something was amiss at Starlink. Its satellites were too big, expensive, and difficult to manufacture. In order to reach a profitable scale, they would have to be made at one-tenth the cost and ten times faster. But the Starlink team did not seem to feel much urgency, a cardinal sin for Musk. So one Sunday night that June, without much warning, he flew to Seattle to fire the entire top Starlink team. He brought with him eight of his most senior SpaceX rocket engineers. None knew much about satellites, but they all knew how to solve engineering problems and apply Musk’s algorithm. The engineer he tapped to take over was Mark Juncosa, who was already in charge of structural engineering at SpaceX. That had the advantage of integrating the design and manufacture of all SpaceX products, from the boosters to the satellites, under one manager. It also had the advantage of that person being Juncosa, a feverishly brilliant engineer who could mind-meld with Musk.

  • When Juncosa took over at Starlink, he threw away the existing design and started back at a first-principles level, questioning every requirement based on fundamental physics. The goal was to make the simplest communications satellite possible, and later add bells and whistles. “We had marathon meetings, and Elon pushed on every little thing,” says Juncosa. For example, the satellite’s antennas were on a separate structure from the flight computer. The engineers had decreed that they be thermally isolated from one another. Juncosa kept asking why. When told that the antennas might overheat, Juncosa asked to see the test data. “By the time that I asked ‘Why?’ five times,” Juncosa says, “people were like, ‘Shit, maybe we should just make this one integrated component.’ ” By the end of the design process, Juncosa had turned a rat’s nest into what was now a simple flat satellite. It had the potential to be an order of magnitude cheaper. More than twice as many could be packed into the nose cone of a Falcon 9, doubling the number each flight could deploy. “I was, like, pretty happy with it,” Juncosa says. “I’m sitting there thinking how clever I had been.” But Musk was still picking over each detail. When they were launched on a Falcon 9, there were connections holding each satellite down so that they could be released one at a time and not bump into each other. “Why not release them all at once?” he asked. That initially struck Juncosa and the other engineers as crazy. They were afraid of collisions. But Musk said the motion of the spaceship would cause them to separate naturally. If they did happen to bump, it would be very slow and harmless. So they got rid of the connectors, saving a little bit of cost, complexity, and mass. “Life got way easier because we culled those parts,” Juncosa says. “I was too chicken to propose that, but Elon made us try it.”

  • By May 2019, the design of the simplified Starlink was complete and the Falcon 9 rocket began launching them into orbit. When they became operational four months later, Musk was at his south Texas house and went on Twitter. “Sending this tweet through space via Starlink satellite,” he wrote. He was now able to tweet on an internet that he owned.

  • Musk was so pleased with the concept of Starship that one afternoon, during a meeting in the SpaceX conference room, he impulsively decided to deploy his burn-the-boats strategy. Cancel the Falcon Heavy, he ordered. The executives in the room texted Gwynne Shotwell what was happening. She rushed from her cubicle, plopped in a chair, and told Musk he could not do that. The Falcon Heavy, with three booster cores, was key to fulfilling its contracts with the military to launch large intelligence satellites. She had the standing to get away with such a challenge. “Once I gave Elon the context, he agreed we couldn’t do what he wanted,” she says. One problem Musk had was that most people around him were afraid to do that.

  • One evening, his young cousin James Musk, a software coder on the Autopilot team, was having dinner with the team’s leader, Milan Kovac, at a fancy San Francisco restaurant when his phone rang. “I saw it was from Elon, and thought, ‘Oh, this is bad,’ ” he remembers. He walked to the parking lot, where he listened to Musk describe in a dark tone how Tesla would go bankrupt if they didn’t do something dramatic. For more than an hour, Musk questioned James about who on the Autopilot team was really good. As often happened when he went into crisis mode, he wanted to clean house and fire people, even in the midst of a surge. He decided he needed to get rid of all the top managers of the Autopilot team, but Omead Afshar intervened to convince him to at least wait until after Autonomy Day. Shivon Zilis, who had the thankless task of trying to serve as a buffer between Musk and the team, also tried to delay the firings, as did Sam Teller. Musk reluctantly agreed to wait until right after Autonomy Day, but he wasn’t happy. He moved Zilis out of Tesla to a job at Neuralink. Teller also ended up leaving during the turmoil. James Musk was given the task of trying to integrate into the Autopilot software the ability to see red and green traffic lights, a pretty basic task that was not yet part of the system. He got it to work pretty well, but it became clear that the team would not be able to meet Musk’s challenge of demonstrating a car that would drive on its own through Palo Alto. As Autonomy Day neared, he scaled his demand back to a task that was merely, as he later described it, “insanely difficult”: the car would drive around the Tesla headquarters, go onto the highway, do a loop involving seven turns, and then return. “We did not believe we could do what he demanded, but he believed we could,” says Anand Swaminathan, a member of the Autopilot team. “In just weeks, we were able to make it do seven difficult turns.” In his Autonomy Day presentation, Musk mixed, as he often did, vision and hype. Even in his own head, he blurred the line between what he believed and what he wanted to believe. Tesla, he said yet again, was within a year of creating a fully autonomous vehicle. At that point the company would deploy a million Robotaxis that people could summon for rides. One day in late 2018, Musk was sitting at his desk at Tesla headquarters in Palo Alto playing with a small toy version of the Model S. It looked like a miniaturized copy of the real car, and when he took it apart he saw that it even had a suspension inside. But the entire underbody of the car had been die cast as one piece of metal. At a meeting of his team that day, Musk pulled out the toy and put it on the white conference room table. “Why can’t we do that?” he asked. One of the engineers pointed out the obvious, that an actual car underbody is much bigger. There were no casting machines to handle something that size. That answer didn’t satisfy Musk. “Go figure out how to do it,” he said. “Ask for a bigger casting machine. It’s not as if that would break the laws of physics.” Both he and his executives called the six major casting companies, five of whom dismissed the concept. But a company called Idra Presse in Italy, which specialized in high-pressure die-casting machines, agreed to take on the challenge of building very large machines that would be able to churn out the entire rear and front underbodies for the Model Y. “We did the world’s largest casting machine,” Afshar says. “It’s a six-thousand-ton one for the Model Y, and we will also use a nine-thousand-ton one for Cybertruck.” The machines inject bursts of molten aluminum into a cold casting mold, which can spit out in just eighty seconds an entire chassis that used to contain more than a hundred parts that had to be welded, riveted, or bonded together. The old process produced gaps, rattles, and leaks. “So it went from a horrible nightmare to something that is crazy cheap and easy and fast,” Musk says. The process reinforced Musk’s appreciation for the toy industry. “They have to produce things very quickly and cheaply without flaws, and manufacture them all by Christmas, or there will be sad faces.” He repeatedly pushed his teams to get ideas from toys, such as robots and Legos. As he walked the floor of the factory, he spoke to a group of machinists about the high-precision molding of Lego pieces. They are accurate and identical to within ten microns, which means any part can easily be replaced by another. Car components needed to be that way. “Precision is not expensive,” he says. “It’s mostly about caring. Do you care to make it precise? Then you can make it precise.”

  • After SpaceX’s launch of astronauts to the Space Station in May 2020, it had an impressive run of eleven unmanned successful satellite launches in five months. But Musk, as always, feared complacency. Unless he maintained a maniacal sense of urgency, he worried, SpaceX could end up flabby and slow, like Boeing. Following one of the launches that October, Musk paid a late-night visit to Pad 39A. There were only two people working. Sights like that triggered him. At all of his companies, as the employees at Twitter would discover, he expected everyone to work with an unrelenting intensity. “We have 783 employees working at the Cape,” he said in a cold rage to his launch VP there. “Why are there only two of them working now?” Musk gave him forty-eight hours to prepare a briefing on what everyone was supposed to be doing. When he didn’t get the answers he wanted, Musk decided to find out for himself. He went into hardcore, all-in mode. As he did at the Nevada and Fremont Tesla factories, and as he would later do at Twitter, he moved into the building, in this case the hangar at Cape Canaveral, and went to work around the clock. His all-night presence was both performative and real.

  • “We need to stack the ship on the booster,” Musk told the impromptu meeting of a hundred workers gathered in a semicircle in one of the three hangar-like tents in Boca Chica. It was a brutally sunny day in July 2021, and he was focused on getting FAA approval for Starship to fly. The best way, he decided, was to stack the booster and the second-stage ship on the launchpad to show that they were ready. “That will force the regulators to get off their butts,” he said. “There will be public pressure getting them to move to approval.” It was a somewhat pointless but typical Musk move. Starship, as it turned out, would not be ready to fly until April 2023, another twenty-one months away. But creating a maniacal sense of urgency would, he hoped, light a fire under everyone, including the regulators, the workers, and even himself. For the next few hours, he lumbered along the assembly lines, his hairless arms swinging, his neck slightly bent, pausing occasionally to stare at something in silence. Increasingly, his face got darker, and his pauses took on an ominous feel. By 9 p.m., a full moon had arisen out of the ocean, and it seemed to be transforming him into a man possessed. I had seen Musk get into this demon-mode temperament before, so I sensed what it portended. As often happens—at least two or three times a year in a major way—a compulsion was swelling inside him to order up a surge, an all-in burst of round-the-clock activity, like he had done at the Nevada battery factory, the Fremont car-assembly plant, and the autonomous-driving team offices, and would later do in the crazed month after he bought Twitter. The goal was to shake things up and “extrude shit out of the system,” as he put it. The storm clouds building in his head burst when he and a group of his top managers went down the road to the launchpad site and didn’t see anybody working. This might not have seemed unusual to most people on a late Friday night, but Musk erupted.

  • By the time Musk got back from the launchpad to the Starbase main building, the video monitor by the front door had been reprogrammed. It read, “Ship+Rocket Stacked T –196h 44m 23s,” and was counting down the seconds. Balajadia explained that Musk does not let them round off into days or even hours. Every second counted. “We need to get to Mars before I die,” he said. “There’s no forcing function for getting us to Mars other than us, and sometimes that means me.” The surge was successful. In just over ten days, the booster and spacecraft of Starship were stacked on the launchpad. It was also a bit pointless. The rocket was not yet capable of flying, and stacking it did not force the FAA to rush its approval. But the ginned-up crisis pushed the team to remain hardcore, and it provided Musk with a bit of the drama that his headspace craves. “I feel renewed faith in the future of humanity,” he said that evening. Another storm had passed.

  • Gwynne Shotwell and the SpaceX CFO, Bret Johnsen, arranged a small meeting one afternoon with the person in the finance department in charge of overseeing Raptor costs. In walked a studious-looking young financial analyst named Lucas Hughes, whose slightly preppy appearance was mitigated by his hair being scrunched into a ponytail. He had never directly interacted with Musk and wasn’t even sure Musk knew his name. So he was nervous. Musk began with his lecture on collegiality. “I want to be super clear,” he began. “You are not the friend of the engineers. You are the judge. If you’re popular among the engineers, this is bad. If you don’t step on toes, I will fire you. Is that clear?” Hughes stuttered a bit as he assented. Ever since he flew back from Russia and calculated the costs of building his own rockets, Musk had deployed what he called the “idiot index.” That was the ratio of the total cost of a component to the cost of its raw materials. Something with a high idiot index—say, a component that cost $1,000 when the aluminum that composed it cost only $100—was likely to have a design that was too complex or a manufacturing process that was too inefficient. As Musk put it, “If the ratio is high, you’re an idiot.” “What are the best parts in Raptor as judged by the idiot index?” Musk asked. “I’m not sure,” Hughes responded. “I will find out.” This was not good. Musk’s face hardened, and Shotwell shot me a worried glance. “You better be fucking sure in the future you know these things off the top of your head,” Musk said. “If you ever come into a meeting and do not know what are the idiot parts, then your resignation will be accepted immediately.” He spoke in a monotone and showed no emotion. “How can you fucking not know what the best and worst parts are?” “I know the cost chart down to the smallest part,” Hughes said quietly. “I just don’t know the cost of the raw materials of those parts.” “What are the worst five parts?” Musk demanded. Hughes looked at his computer to see if he could calculate an answer. “NO! Don’t look at your screen,” Musk said. “Just name one. You should know the problematic parts.” “There’s the half nozzle jacket,” Hughes offered tentatively. “I think it costs thirteen thousand dollars.” “It’s made of a single piece of steel,” Musk said, now quizzing him. “How much does that material cost?” “I think a few thousand dollars?” replied Hughes. Musk knew the answer. “No. It’s just steel. It’s about two hundred bucks. You have very badly failed. If you don’t improve, your resignation will be accepted. This meeting is over. Done.”

  • When Hughes came into the conference room the next day for a follow-up presentation, Musk showed no sign that he remembered reaming him out. “We are looking at the twenty worst ‘idiot index’ parts,” Hughes began as he pulled up a slide. “There’s definitely some themes.” Other than wringing a pencil, he was able to hide his nervousness. Musk listened quietly and nodded. “It’s mainly the parts that require a lot of high-precision machining, like pumps and fairings,” Hughes continued. “We need to cut out as much of the machining as possible.” Musk started smiling. This had been one of his themes. He asked a few specific questions about the use of copper and the best way to do stamping and hole-punching. It was no longer a quiz or a confrontation. Musk was interested in figuring out the answers. “We are looking at some of the techniques that automakers use to keep these costs down,” Hughes continued. He also had a slide that showed how they were applying Musk’s algorithm to each of the parts. There were columns that showed what requirements had been questioned, what parts had been deleted, and the name of the specific person in charge of each component. “We should ask each of them to see if they can get the cost of their part down by eighty percent,” Musk suggested, “and if they can’t, we should consider asking them to step aside if someone else might be able to do so.” By the end of the meeting, they had a roadmap to get the cost of each engine down from $2 million to $200,000 in twelve months.

  • After a month, just as suddenly as he had forced his team to focus on a futurist 1337 engine, Musk turned their attention back to revising the current Raptor engine into a leaner and meaner Raptor 2. “I’m moving the SpaceX propulsion focus back to Raptor,” he announced in a 2 a.m. text message. “We need engine production rate to be one per day to keep up a decent launch cadence. It is currently at one every three days.” I asked whether that would slow the development of the 1337. “Yes,” he replied. “We can’t make life multiplanetary with Raptor, as it is way too expensive, but Raptor is needed to tide us over until 1337 is ready.” Was the 1337 surge and retreat a carefully considered strategy by Musk to get his team thinking more boldly, or was it an impulsive act that he later walked back? As usual with him, it was a mix of both. It served a purpose of forcing new ideas, including getting rid of various shrouds and skirts, that would be incorporated into his goals for an improved Raptor. “The exercise helped define what an ideal engine looks like,” McKenzie says. “But it was not moving the needle on stuff that was immediately needed to advance the Starship program.” Over the next year, McKenzie and his team were able to churn out Raptors almost like they were cars on an assembly line. By Thanksgiving 2022, they were making more than one a day, creating a stockpile for future Starship launches.

  • A few months later, Musk came by the Neuralink lab in Fremont, near the Tesla factory, where the engineers showed him their latest version. It combined four separate chips, each with about a thousand threads. They would be implanted in different parts of the skull with wires connecting them to a router that was embedded behind the ear. Musk paused silently for almost two minutes, while Zilis and her colleagues watched. Then he delivered his verdict: he hated it. It was too complex, with too many wires and connections. He was in the process of deleting connections from SpaceX’s Raptor engines. Each was a possible failure point. “This has to be a single device,” he told the deflated Neuralink engineers. “A single elegant package with no wires, no connections, no router.” There was no law of physics—no basic principle—that prevented all of the functionality from being on one device. When the engineers tried to explain the need for the router, Musk’s face turned stony. “Delete,” he said. “Delete, delete, delete.” After they left the meeting, the engineers went through the usual stages of post-Musk distress disorder: baffled, then angry, then anxious. But within a week they got to the stage of being intrigued, because the new approach, they realized, might actually work. When Musk returned to the lab a few weeks later, they showed him a single chip that could handle the processing of data from all the threads and transmit it by Bluetooth to a computer. No connections, no router, no wires. “We thought this was impossible,” one of the engineers said, “but now we’re actually pretty stoked by it.” One problem they faced was caused by the requirement that the chip be very small. That made it a challenge for it to have a long battery life and support many threads. “Why does it have to be so small?” Musk asked. Someone made the mistake of saying that it was one of the requirements they had been given. This flipped on the switch for Musk to intone his algorithm, beginning with questioning every requirement. Then he engaged them on the basic science of the chip size. Our skulls are rounded, so couldn’t the chip bulge a bit? And couldn’t the diameter be bigger? They came to the conclusion that a human skull could easily accommodate a larger chip. When they had the new device ready, they implanted it in one of the macaque monkeys, named Pager, housed at the lab. He was taught to play the video game Pong by rewarding him with a fruit smoothie when he scored well. The Neuralink device recorded which neurons were firing each time he moved the joystick a certain way. Then the joystick was deactivated, and the signals from the monkey’s brain controlled the game. It was a major step toward Musk’s goal of creating a direct connection between a brain and a machine. Neuralink uploaded a video of it to YouTube, and within a year it was viewed six million times.

  • At the meeting, Musk found Agrawal to be likable. “He’s a really nice guy,” he says. But that was the problem. If you ask Musk what are the traits needed in a CEO, he would not include “being a really nice guy.” One of his maxims is that managers should not aim to be liked. “What Twitter needs is a fire-breathing dragon,” he said after that meeting, “and Parag is not that.”

  • The twenty engineers who were scheduled to be part of the presentation huddled in a backstage area telling war stories. Phil Duan, a young machine-learning expert on the Autopilot team, studied optic information science in his hometown of Wuhan, China, then got a PhD at Ohio University. He joined Tesla in 2017, just in time for the crazed surges that culminated with Musk’s push to unveil a self-driving car on Autonomy Day in 2019. “I worked for months without a day off and got so tired that I quit Tesla right after Autonomy Day,” he said. “I was burned out. But after nine months, I was bored, so I called my boss and begged him to let me come back. I decided I’d rather be burned out than bored.” Tim Zaman, who led the artificial intelligence infrastructure team, had a similar story. From northern Holland, he joined Tesla in 2019. “When you’re at Tesla, you’re afraid to go anywhere else, because you will become so bored.” He just had his first child, a daughter, and knows that Tesla is not conducive to a work-life balance. Nevertheless, he plans to stay. “I’m going to spend the next few days off with my wife and daughter,” he says, “but if I take a whole week off, my brain gets fried.”

  • The issue was not merely the facilities. Between Twitterland and the Muskverse was a radical divergence in outlook that reflected two different mindsets about the American workplace. Twitter prided itself on being a friendly place where coddling was considered a virtue. “We were definitely very high-empathy, very caring about inclusion and diversity; everyone needs to feel safe here,” says Leslie Berland, who was chief marketing and people officer until she was fired by Musk. The company had instituted a permanent work-from-home option and allowed a mental “day of rest” each month. One of the commonly used buzzwords at the company was “psychological safety.” Care was taken not to discomfort. Musk let loose a bitter laugh when he heard the phrase “psychological safety.” It made him recoil. He considered it to be the enemy of urgency, progress, orbital velocity. His preferred buzzword was “hardcore.” Discomfort, he believed, was a good thing. It was a weapon against the scourge of complacency. Vacations, flower-smelling, work-life balance, and days of “mental rest” were not his thing. Let that sink in.

  • Shroff showed Musk instances where a neural network planner would work better than a rules-based approach. The demo had a road littered with trash cans, fallen traffic cones, and random debris. A car guided by the neural network planner was able to skitter around the obstacles, crossing the lane lines and breaking some rules as necessary. “Here’s what happens when we move from rules-based to network-path-based,” Shroff told him. “The car will never get into a collision if you turn this thing on, even in unstructured environments.” It was the type of leap into the future that excited Musk. “We should do a James Bond–style demonstration,” he said, “where there are bombs exploding on all sides and a UFO is falling from the sky while the car speeds through without hitting anything.” Machine-learning systems generally need a goal or metric that guides them as they train themselves. Musk, who liked to manage by decreeing what metrics should be paramount, gave them their lodestar: the number of miles that cars with Tesla Full Self-Driving were able to travel without a human intervening. “I want the latest data on miles per intervention to be the starting slide at each of our meetings,” he decreed. “If we’re training AI, what do we optimize? The answer is higher miles between interventions.” He told them to make it like a video game where they could see their score every day. “Video games without a score are boring, so it will be motivating to watch each day as the miles per intervention increases.” Members of the team installed massive eighty-five-inch television monitors in their workspace that displayed in real time how many miles the FSD cars were driving on average without interventions. Whenever they would see a type of intervention recurring—such as drivers grabbing the wheel during a lane change or a merge or a turn into a complex intersection—they would work with both the rules and the neural network planner to make a fix. They put a gong near their desks, and whenever they successfully solved a problem causing an intervention, they got to bang the gong.