Part IThe Story
When You Shut One Eye
The friend is never named. Neither is the car, the road, or the hour. What survives, in the version Amnon Shashua gave to the Weizmann Institute's alumni magazine in the summer of 2014, is one gesture and one sentence. A young man, recently out of a master's program, sits in a car beside a friend, and his head is full of what he has spent two years absorbing in Shimon Ullman's laboratory in Rehovot: how the brain turns light into objects, and how a machine might be taught to do the same. He points at the windshield. "We could put a camera right here in the car to give us a collision warning."
Nothing happens. The remark lodges, in the magazine's phrase, "in the back of his head," and stays there for years. Late-1980s Israel had no automotive industry to speak of, the computers of the period could barely hold a single image in memory, and the young man was about to leave for Cambridge, Massachusetts, to study the geometry of seeing at MIT.
Years later the sentence acquired a sequel. Shashua, by then a professor, was presenting to a room of car-industry engineers, and one of them asked about a collision-warning system that used two cameras. The question was reasonable. It was, in fact, the orthodoxy. Depth perception, as every engineer in that room had been taught, requires two viewpoints. Stereo vision triangulates; the gap between two lenses supplies the baseline from which distance is computed, the same way the gap between a person's eyes does. The automakers who had begun experimenting with driver assistance were working with radar emitters and stereoscopic rigs, expensive hardware that measured the world precisely and therefore lived only in luxury cars.
— Amnon Shashua, recalling his answer to automotive engineers (Weizmann Institute, 2014)You don't need two cameras. One is enough. When you shut one eye, you don't go blind.
It sounds like a quip. It was a research program, a business plan, and a philosophical position compressed into three short sentences, and nearly everything Shashua built over the following four decades can be read as an elaboration of it. A single eye cannot triangulate, but it can watch. An approaching car grows in the frame; a receding one shrinks; the rate of that change, measured against time, yields distance and closing speed without a second lens. The system Mobileye eventually shipped did exactly this, comparing the relative size and position of objects in front of the car every 27 milliseconds and computing trajectories from the differences. The camera was cheap. The mathematics was expensive. And mathematics, once written, costs almost nothing to copy into the next car, and the next ten million.
The wager made him rich, very rich, and then publicly, uncomfortably mortal. It produced the largest Israeli IPO in American history in 2014 and the largest acquisition of an Israeli technology company in 2017, when Intel paid $15.3 billion for a firm that, according to the Süddeutsche Zeitung, had only a few hundred employees. It put his chips into more than 55 million cars by 2020. It also placed him, in the summer of 2016, at the center of an argument about what a camera had failed to see on a Florida highway, and in the summer of 2024, at the center of a different argument, about what a company had failed to foresee in its order book. On July 23, 2026, at 66, he announced he would step down as chief executive of the company he had founded 27 years earlier.
"There is never a good time to retire," he said on the conference call. He did not say anything about eyes.
By the Numbers
The Monocular Bet
$15.3BIntel's price for Mobileye in 2017, then the largest acquisition of an Israeli tech company
$5.3BMarket capitalization at the August 2014 NYSE debut, the largest Israeli IPO in the U.S.
27 msInterval at which the single-camera system recalculates the size and position of objects ahead
55M+Vehicles carrying Mobileye technology by 2020
6Companies co-founded around applied AI, from Mobileye to AAI Technologies
~$60MDonated by the Shashua family to Israeli causes as of 2024
4,200Mobileye employees at the time of his July 2026 resignation
A Day a Week
Shimon Ullman did not tell his student to start a company. He showed him how.
Ullman is a vision scientist of the old Weizmann school, a man whose career runs along the seam between what the brain does and what a computer can be made to do, and he was, in the 1980s, already proving that the seam could be monetized without being torn. He co-founded Orbot, the company that later became Orbotech, which built automated optical inspection systems for the printed-circuit industry: machines that looked at circuit boards and found the flaws a human inspector would miss. He kept his laboratory. He kept his chair. In 2014 he received the EMET Prize, one of Israel's grand academic honors. He was, in short, both things at once, and he made the arrangement look ordinary.
"Without the knowledge gained in Shimon Ullman's lab and watching him work the way he did — always thinking about how his lab insights might help make life better for people — I wouldn't be where I am today," Shashua said in 2014, from Mobileye's Jerusalem headquarters. When he finally had a company of his own to run, he copied his teacher's calendar precisely: one day a week at the company, four days at the university.
That ratio would invert, eventually, and then invert again, and the question of which of the two lives was the real one would hover over him until the morning he resigned. But in the beginning it was a clean formula, and it was borrowed.
The first paper bearing his name, presented at the Second International Conference on Computer Vision in 1988, is credited to "S. Ullman and A. Sha'ashua," the surname transliterated with an apostrophe that would later quietly disappear. Its title was "Structural Saliency: The Detection of Globally Salient Structures using a Locally Connected Network." The problem it addressed is one every animal solves without thinking: in a field of visual clutter, which shapes matter? Which curves cohere into an object, and which are noise? The work, Shashua wrote decades later, had been inspired by Anne Treisman, the psychologist whose experiments on attention showed that the mind does not see everything at once but binds features together selectively. "Since my days as a student," he wrote in 2022, "brain science has been a subject very close to my heart."
It is a revealing first problem for a man who would spend his life teaching cars to look at roads. Saliency is the question of what to ignore.
He had come to it from Ramat Gan, the dense city just east of Tel Aviv, born on May 26, 1960, to parents who had come from Iraq. The Mizrahi immigrants of that generation arrived in a state built largely by and for European Jews, and their children grew up in its hinterlands of opportunity, close to the center geographically and further from it in every other way. The boy was fascinated by computers at a moment when the field barely existed, and he went to the ORT school in neighboring Givatayim, part of the vocational network that had spent a century training Jewish children for the trades of whatever century they happened to be living in. He took a bachelor's degree in mathematics and computer science at Tel Aviv University in 1985, and then went to Ullman.
The ordinary path from there, for a brilliant young Israeli working on machines that perceive and navigate, ran through the military. Joachim Becker of the Süddeutsche Zeitung made the point drily in a 2018 profile: anyone in Israel working on computer brains that recognize their surroundings could make a career in the army. "Doch Shashua promovierte in den USA." But Shashua took his doctorate in the United States.
He did it at MIT, in the Department of Brain and Cognitive Sciences, while doing the actual work at the Artificial Intelligence Laboratory, which is to say he was enrolled in the study of minds and employed in the study of machines. His 1993 dissertation was titled Geometry and Photometry in 3D Visual Recognition: how an object's shape and the light falling on it combine to make it recognizable from a flat image. He stayed on as a postdoctoral fellow under Tomaso Poggio, the Italian-born physicist turned computational neuroscientist whose career has been one long argument that intelligence should be studied as a single problem across biology and engineering. Poggio's center, then called the Center for Biological and Computational Learning, later became the Center for Brains, Minds and Machines. In March 2017, two weeks after Intel announced it would pay $15.3 billion for Mobileye, Shashua came back to that center to lecture, an alumnus returning to the room where the idea had been sharpened.
In 1996 he went home and joined the computer science faculty of the Hebrew University of Jerusalem. He never left it. Associate professor in 1999, the year he founded Mobileye. Full professor in 2003. Head of the School of Engineering and Computer Science from 2002 to 2005, years when Mobileye was still burning through private capital and had not yet shipped a chip. The Sachs Chair in Computer Science in 2007. Through the IPO, through Intel, through the second IPO and the slide that followed, the title on the bios has always been Prof. Amnon Shashua. Even Mobileye's investor-relations page leads with it.
The Patent Hollywood Didn't Want
The first thing he tried to sell, nobody bought.
It was a patent for an algorithm that allowed easy editing of light in images: take a photograph of a face, and re-render it as though the light had come from somewhere else. The academic version would be published in 2001, with his student Tammy Riklin-Raviv, as "The Quotient Image: Class-Based Re-Rendering and Recognition with Varying Illuminations." Shashua pitched it to Hollywood studios. They passed. A Japanese company eventually adopted it for the fashion industry, an outcome so oblique to the original pitch that it reads like a parable about the gap between what a scientist thinks his invention is for and what a market decides it is for.
"Initially," the Weizmann profile notes, with the restraint of an alumni publication, "the transition from developing scientifically proven algorithms to generating marketable products was more challenging than he had imagined."
His second patent was for a method of turning two-dimensional images into three-dimensional models, and it found a buyer because he found the right problem for it. In 1995, a year before he joined the Hebrew University faculty, he founded CogniTens, which used the technique for quality control in the automotive spare-parts industry. The product photographed a manufactured part, reconstructed it in three dimensions, compared the reconstruction against the original design, and flagged the deviations. It was Orbotech's logic, Ullman's logic, transplanted from circuit boards to car parts: a machine that looks at a manufactured object and notices what is wrong with it. Calcalist later reported that he was also involved, in the same period, in founding CogniTech, which applied image processing and video analysis to forensic investigation and identification.
The Weizmann account calls CogniTens "his first encounter with strategic product-oriented thinking," and says he enjoyed it. That is the most important sentence in his early biography, more important than any of the prizes. Plenty of brilliant computer scientists have tried business and returned from it bruised and relieved. Shashua found he liked the problem of deciding what to build.
CogniTens was sold to the Swedish measurement company Hexagon AB in 2007. By then he had already started its successor, and the lesson of the first venture, buried in its customer list, had shaped the second: the automotive industry, conservative and procurement-obsessed and terrified of liability, would buy an idea from an Israeli professor if the idea saved it money and could be proven in a lab.
There is one more episode from these years that rarely appears in the corporate biographies. In the mid-2000s, according to the record of his research output, Shashua contributed to the Golem Group's entry in the DARPA Grand Challenge, the Pentagon-funded desert races that functioned as the founding myth of the American self-driving car. The famous names of that era were Stanford and Carnegie Mellon, and the famous alumni went on to Google's car project and its descendants. Shashua was in the race, in a minor key, and then went back to building something much less glamorous: a camera that warned you when you were drifting out of your lane.
Cheap Glass, Expensive Mathematics
Mobileye was incorporated in Jerusalem on May 13, 1999. Three founders appear in the record: Shashua; Ziv Aviram, the businessman who would run the company's operations and serve for years as its chief executive; and Norio Ichihashi, a Japanese name the public record barely annotates. Aviram, a manager rather than a scientist, was the other half of the arrangement Shashua had copied from Ullman. The professor would supply the eye. Someone else would supply the hands.
The thesis, as Calcalist summarized it decades later, was "that a single camera combined with advanced computer vision algorithms could help prevent traffic accidents." Mobileye's own corporate history phrases it more like a margin structure: adding "intelligence" to inexpensive cameras. Both are accurate. The second is more revealing.
The German automakers, Becker wrote in the Süddeutsche, said it was impossible. Unmöglich. With their radar systems they could measure the world to the centimeter. What they could not do was understand what was happening around the car. Radar knows that something is forty meters away and closing. It does not know whether that something is a truck, a pedestrian, a cardboard box, or a bridge overhead. A camera, if you can make it think, knows what a thing is. Shashua was betting that classification would matter more than range, and that the expensive sensor would lose to the cheap one plus a decade of mathematics.
The decade was the hard part. The company opened its first research center in 2004. That same year, its engineers published two papers at the IEEE Intelligent Vehicles Symposium that read, in retrospect, like a manifesto stated in the passive voice. One, by Erez Dagan, Ofer Mano, Gideon Stein, and Shashua, was titled "Forward Collision Warning with a Single Camera." The other, by Shashua, Yoram Gdalyahu, and Gaby Hayon, was "Pedestrian Detection for Driving Assistance Systems: Single-Frame Classification and System Level Performance." Single camera. Single frame. The word kept recurring, as though repetition might wear down the skeptics.
Those names would recur too. When the European Patent Office named Shashua a finalist for its European Inventor Award in 2019, it named the team alongside him: Erez Dagan, Yoram Gedalyahu, Gaby Hayon, Elchanan Rushinek, Shai Shalev-Shwartz, and Gideon Stein. Fifteen years on, the same handful of people. The company that would be sold for $15.3 billion had, at its core, the stability of a research group.
The key commercial decision was not algorithmic. Mobileye did not sell software to run on someone else's processor. It designed its own: a system-on-chip called EyeQ, with the vision algorithms built into the silicon. EyeQ1 reached the market in 2008, fabricated on a 180-nanometer process, drawing about 2.5 watts, and delivering roughly 0.0044 trillion operations per second, a figure so small that a modern phone would not notice it. That chip went into a Continental camera in the fifth-generation BMW 7 Series. General Motors launched Mobileye-powered lane departure warning on the Cadillac STS and DTS. EyeQ2 followed in 2010, EyeQ3 in 2014, EyeQ4 in 2018, each a process node smaller, each running a longer list of neural networks: vehicles' rears, pedestrians, lane markings, traffic signs, then traffic lights and road edges and animals and the trajectories of other road users.
Owning the chip meant owning the bill of materials. A carmaker could not easily swap out Mobileye's software for a competitor's because the software and the hardware were the same product. And because the hardware was a cheap camera plus a cheap chip rather than a radar array, driver assistance stopped being a luxury option. The EPO's citation put it plainly: ADAS had "entered the market using cumbersome and expensive sensors" that "restricted their use to high-end vehicles," and Mobileye "reversed this trend."
In March 2019, irritated by a rival, Shashua published his own list of the company's firsts, and it reads like a geological record of the monocular thesis advancing one stratum at a time: camera and radar fusion in 2007; pedestrian-detection warning in 2010; camera-only forward-collision warning in 2011; camera-only adaptive cruise control in 2013; hands-free assist in 2015; crowd-sourced HD mapping in 2016. Camera-only. Each entry was a sensor the industry had thought necessary, subtracted.
The money followed the subtraction. In July 2013, Mobileye sold a 25 percent stake to a group of blue-chip investors for $400 million, a valuation of about $1.5 billion. On August 1, 2014, it listed on the New York Stock Exchange, raising $890 million at a market capitalization of $5.3 billion: the largest IPO of an Israeli company in American history. By the end of that year its technology was in 160 car models built by 18 manufacturers. By February 2016, Shashua's own university biography counted "approximately 10 million cars from 23 automobile manufacturers." By 2019, the EPO counted more than 40 million. By 2020, the Dan David Prize citation counted more than 55 million.
That February 2016 biography contains one other sentence, written by the professor about his own company, which time would make very hard to read. "An early version of Mobileye's autonomous driving technology was deployed in series as an 'autopilot' feature in October, 2015."
Lateral Turn Across Path
On May 7, 2016, on a divided highway in Florida, a Tesla Model S traveling with Autopilot engaged drove into the side of an 18-wheel tractor-trailer that had turned left across its path. The driver was killed. According to the preliminary report of the National Transportation Safety Board, the car was speeding, and both Autosteer lane-keeping and traffic-aware cruise control were active. Tesla's explanation, which became one of the most quoted sentences in the history of the self-driving car, was that neither Autopilot nor the driver had noticed the white side of the trailer against a brightly lit sky, so the brakes were never applied.
The camera behind that windshield was running a Mobileye EyeQ3.
Tesla had announced in August 2015 that it would incorporate Mobileye's technology into the Model S; the hardware, retrospectively called Autopilot Hardware 1, shipped in Model S and Model X cars from September 2014 to October 2016. It was the most visible design win Mobileye had ever had, a Silicon Valley halo on an automotive supplier, and the October 2015 software release was the "autopilot" feature Shashua had mentioned with evident pride in his own biography.
Mobileye's first statement after the crash became public was precise and limiting: its system, as then designed, did not recognize a "lateral turn across path." A truck turning across the car's lane, seen broadside, was not a scenario the EyeQ3 generation had been built to classify for automatic braking. The software had been designed to recognize the rear of a vehicle, as the 2004 papers had specified. This one was showing its side.
In July 2016, Mobileye announced it would end its relationship with Tesla. Reports at the time said Tesla had not shared its plans with Mobileye. In September, Shashua, then Mobileye's chief technology officer and executive chairman, with Aviram as chief executive, said in public what the company's statements had only implied. Tesla, he told Reuters, "was pushing the envelope in terms of safety." The system, he said, "is not designed to cover all possible crash situations in a safe manner."
— Amnon Shashua, to Reuters, September 2016No matter how you spin it, [Autopilot] is not designed for that. It is a driver assistance system and not a driverless system.
The rupture is usually told as a clash of temperaments: the cautious Israeli professor and the Californian who named a lane-keeping system after an aircraft feature. That version is true and incomplete. The more interesting reading is that the professor had been caught inside his own metaphor. One eye is enough: enough to warn, enough to brake for the back of a car, enough to keep a lane on a highway. One eye is not enough for a driver who has stopped looking, and Tesla's customers, as their own YouTube videos showed, had stopped looking. The technology had worked as specified. The specification had been put into a product whose name invited people to believe it was something else.
The week Shashua spoke to Reuters, Tesla announced that its new software would use the car's radar "as a primary control sensor without requiring the camera to confirm visual image recognition." The company that had been Mobileye's most famous camera customer was, in its first act after the divorce, demoting the camera.
Years later the two companies would reverse positions with comic symmetry. Tesla would go all-in on cameras, abandoning radar entirely. Mobileye would build its robotaxi system around 13 cameras, three long-range lidars, six short-range lidars, and six radars, and call the philosophy "True Redundancy." The monocular evangelist had become the man insisting on a second opinion.
Formalizing Caution
What do you do when a machine you built is blamed for a death it was never designed to prevent? An engineer answers by improving the machine. A scientist answers by redefining the question.
In 2017, Shashua, Shai Shalev-Shwartz, and Shaked Shammah posted a paper to arXiv titled "On a Formal Model of Safe and Scalable Self-Driving Cars." It introduced a framework they called Responsibility-Sensitive Safety, RSS, which attempted something that sounds either visionary or absurd depending on your temperament: to write down, as mathematics, what it means to drive carefully.
Shalev-Shwartz is the co-author whose name sits beside Shashua's on nearly everything that matters after 2016, and his biography rhymes with his mentor's almost line for line. A professor at the Hebrew University's Rachel and Selim Benin School of Computer Science and Engineering, he spent time as a research assistant professor at the Toyota Technological Institute in Chicago and worked in research at Google and IBM before coming home. In 2014 he co-wrote Understanding Machine Learning: From Theory to Algorithms, which became one of the standard university texts on the subject. Mobileye's own description says he is drawn to autonomous driving "as the first large scale deployment of artificial intelligence and machine learning outside of the cybernetic world." He became Mobileye's chief technology officer in 2019. He is a theorist who wanted his theorems to have consequences in traffic, which is to say he is a younger Shashua, and Shashua was a younger Ullman.
The intuition behind RSS is old and human. If you rear-end someone, it is your fault, because you were supposed to leave enough room to stop. RSS generalizes that: it defines, for every geometry of road situation, a safe longitudinal and lateral distance, given the vehicles' speeds and the worst-case reaction time and braking power; it defines who cut in on whom and therefore who bears responsibility for restoring the gap; it defines how right of way enters the calculation; and it specifies how to behave cautiously when the sensors cannot see, when a child might step out from behind a parked car. A car that never violates these constraints, the paper argued, will never cause an accident.
Shashua had done the empirical homework. He told the Süddeutsche that he had analyzed six million accidents and that 99.4 percent of them fell into one of 37 typical scenarios. "Our mathematical models cover all of them." Becker, characteristically, described him as a mathematician searching for a Weltformel, a world formula, for automated driving. But the same interview contains the sentence that complicates the formula, and Shashua delivered it, Becker wrote, like a traffic psychologist: calmly, firmly. Driving, he said, is a game with many players. The other cars are not autonomous vehicles and do not necessarily cooperate. Behind their steering wheels sit people who are aggressive and make mistakes.
That is the whole tension of his career in two consecutive thoughts. The world is chaotic, full of aggressive and erring humans. The world can be reduced to 37 scenarios.
In a November 2017 essay for Calcalist, Shashua answered the critics who said RSS was a liability shield dressed up as ethics. "RSS formalizes the common sense of human judgment under a comprehensive set of road situations," he wrote. "With human drivers, the interpretation of responsibility for collisions and other incidents is fluid." Blame is assigned "based on imperfect information and other factors interpreted after the fact." Machines are different: they "have highly accurate information about the environment around them, always know their reaction time and braking power, and are never distracted or impaired." The model, he explained, permits a car to break the law, crossing a solid white line, say, to escape a collision caused by someone else, so long as the escape does not cause a separate accident. What it forbids is trading one crash for a supposedly lesser one. The reason he gave is the most humane sentence in the whole technical literature of Mobileye: "the judgment of accident severity is subjective and might miss hidden, critical variables, such as a baby in the back seat of the seemingly 'less severe' accident."
There is something characteristic in that refusal. The famous trolley-problem debates of the period, about whether a self-driving car should swerve to kill one person rather than five, assumed the machine could know who was in the other car. Shashua's answer was that it could not, and that a safety system premised on knowledge it does not have is not a safety system at all. His formalism is, at bottom, an epistemology of humility: specify only what you can measure.
He gave it away. Mobileye "openly shared all the technical details and mathematics behind RSS," he wrote in a March 2019 editorial distributed by Intel, "because we believe that the safety of automated vehicles should not be proprietary." Baidu and Valeo signed on. China's ITS standards body formed a working group to standardize it. The generosity had a sharp edge, which became visible in the same editorial, titled "Innovation Requires Originality." Nvidia had just announced its own safety framework, the Safety Force Field, and Shashua did not like it. "It is said that imitation is the sincerest form of flattery," he wrote, "and our innovations have not gone unnoticed." Then, plainly: "One industry player in particular habitually follows our lead, and today I would like to set the record straight on its latest imitation." He enumerated the offenses: Mobileye announced crowdsourced mapping in 2016, Nvidia the following year; Mobileye coined "L2+" in 2017, Nvidia announced an L2+ offering in 2019.
The professor who had published his mathematics so that everyone could use it was furious that someone had used it. It is not a contradiction, exactly. Academics give their work away precisely so that it will be cited.
The same formal temperament governed his engineering arguments for the next decade. When Tesla signaled in 2023 that it would move to an "end-to-end" approach, a single neural network taking in pixels and emitting steering commands, "chatGPT style," Shashua and Shalev-Shwartz published a rebuttal arguing the approach was "neither necessary nor sufficient." Data-driven methods were crucial, they wrote, but had to be "carefully embedded within a well-engineered architecture," for three reasons: transparency, so regulators could dictate exact behavior; controllability, because "there is no way to guarantee 'no lapse of judgement' for a fully neuronal system"; and performance. "Even the most advanced LLMs make embarrassing mistakes," Shashua wrote. "Will we trust them to make safety-critical decisions?"
He applied the same discipline to his product roadmap. Mobileye's development philosophy, he explained in January 2023, required "an incremental and modular transition" from one level of autonomy to the next. "Otherwise, development becomes a series of 'moonshots' where the development and validation of one blade is discarded and a completely new development and validation is required for the next." The heavy lifting would happen once, in a hands-off, eyes-on system called SuperVision, running "solely on cameras." Lidar and radar would be added later as redundancy, to take the driver's eyes off the road. One eye first. Then, only when the stakes demanded it, a second opinion.
The Addendum
On March 12, 2017, the day before Intel announced it would buy Mobileye, Shashua signed a two-page document that reveals more about him than any interview he has given.
It was an addendum to his employment agreement, dated July 24, 2014, the week before the IPO. In the normal course of an acquisition, an executive's unvested stock options and restricted stock units accelerate at closing; the founder becomes liquid on day one and is free to leave on day two. Shashua's addendum waived that. "All outstanding and unvested Company Options and Company RSUs held by Executive as of immediately prior to the Closing shall not accelerate." Instead: no vesting at all before the third anniversary of closing; half on the third anniversary; half on the fourth. If he resigned before then, he forfeited.
The same document contained a second clause, quieter and more pointed. After closing, "the Executive shall report directly to the Chief Executive Officer of the Purchaser." If Intel ever changed that reporting line, if it ever tried to slot the founder of Mobileye beneath some divisional vice president in Santa Clara, the arrangement would count as "Deemed Dismissal," and all his unvested equity would vest immediately.
Read together, the two clauses are a portrait. The first binds him to the company for four years. The second guarantees that the company he is bound to will remain, functionally, his. He gave up liquidity in exchange for altitude. The deal closed in August 2017; Shashua became a senior vice president of Intel and a member of its executive leadership team, as well as president and chief executive of Mobileye. The Süddeutsche summarized his new position bluntly: he was now Intel's boss for everything having to do with autonomous driving.
The price was $15.3 billion. That spring, on Israeli Independence Day, Shashua was chosen to light one of the torches at the national ceremony on Mount Herzl in Jerusalem, an honor that in Israel places a person, briefly, inside the national story rather than merely the business pages. The ORT boy from Ramat Gan, son of Iraqi immigrants, had built the most valuable thing the country's technology industry had ever sold, and the sale had been arranged so that the thing would stay in Jerusalem.
Not every consequence of the deal was ceremonial. Within weeks, the U.S. Securities and Exchange Commission charged two Israelis, Ariel Darvasi and Amir Waldman, with insider trading ahead of the announcement; both had connections to Mobileye through the Hebrew University, a public institution whose scientific community had, by Reuters' count, produced at least 11 Mobileye directors and officers. An Israeli professor settled. A second SEC case, against two Virginia residents who allegedly made about $1 million on the news, unspooled into something closer to a social novel. The trades had been placed in an account dormant since 2011 that suddenly bought Mobileye stock on January 30, 2017. The alleged tip had passed through a brother in Israel, a physician specializing in nonsurgical cosmetic procedures, botox and laser hair removal, whose clinic, the SEC noted, had treated a Mobileye director and his wife. The tipster had allegedly told his friend that his family knew Mobileye's founders and that they were recommending friends and family invest. Neither Intel nor Mobileye was accused of any wrongdoing.
The episode is a footnote, and it is a revealing one. Israel's technology economy is a few universities, a few army units, a few neighborhoods. Its greatest exit was leaked, allegedly, through a dermatology waiting room.
Intel's ownership turned out to be an interlude rather than an ending. In 2020 Intel bought Moovit, the transit-data company, for $900 million to strengthen Mobileye's mobility-as-a-service ambitions. In December 2021 it announced it would take Mobileye public again. On October 26, 2022, Mobileye began trading on Nasdaq under the ticker MBLY; Intel sold 41 million shares for $861 million, and the stock rose more than 37 percent on its first day. The valuation, around $17 billion at pricing, was higher than what Intel had paid five years earlier, which made the round trip look, briefly, like financial genius. Intel kept all the Class B shares and with them 99.4 percent of the voting power. Shashua was chief executive of a public company again, and the company, again, was not quite his.
A Laboratory of Crazy Ventures
Calcalist's Sophie Shulman once described the operation around him as Shashua's "crazy ventures lab," and the phrase is affectionate and accurate. Mobileye's investor-relations page says he has founded six companies "using applied AI in various fields from automotive to assisted wearables to fintech." The list does not include CogniTens. The count, like most of the counts in his biographies, keeps changing.
The second was OrCam, co-founded in 2010, and it is the venture in which the monocular thesis turns most literally human. OrCam's MyEye is a small camera that clips onto a pair of eyeglasses and reads text aloud to people who are blind or visually impaired; it recognizes faces, products, banknotes. TIME named it one of the 100 best inventions of 2019. The technical lineage is plain: the same algorithms that let a single lens understand a road could let a single lens understand a page. The emotional lineage is plainer. A man who had built a career on the claim that closing one eye does not make you blind built a device for people whose eyes had closed.
Then came the language company, AI21 Labs, co-founded in 2017, the year of the Intel sale, long before most of the world had heard the phrase "large language model." It raised $35 million in 2020 and $155 million in a Series C in August 2023 with Google and Nvidia among the investors, the same Nvidia whose imitations he had so publicly resented. Then One Zero, Israel's first digital bank, which won its full license in January 2022, promising ordinary customers the kind of personalized banking previously reserved for the wealthy. Then Mentee Robotics, co-founded in 2022 with Shalev-Shwartz and Lior Wolf, to build humanoid robots. Then AAI Technologies, founded with Shalev-Shwartz, Yoav Levine, Or Sharir, Noam Weiss, and Gal Benyamini, which develops what it calls expert-based AI and whose stated mission, per Mobileye's own biography of its CEO, is "to crack the code of superintelligence." In September 2025, it raised hundreds of millions of dollars at a valuation above $1 billion.
Look at those co-founder lists and the same pattern emerges that ran through the EPO citation. Lior Wolf co-wrote a 2003 paper with Shashua on learning over sets. Or Sharir co-wrote, with Nadav Cohen and Shashua, the 2015 paper "On the Expressive Power of Deep Learning: A Tensor Analysis," part of a theoretical program that brought tensor mathematics to the question of why neural networks work, work that eventually appeared, the Dan David Prize citation noted with evident delight, at top theoretical physics venues. The ventures are staffed by his students and his co-authors. The lab and the company are not two institutions. They are one institution with two letterheads.
Shulman's August 2024 assessment of the empire was titled "The highs and lows of Amnon Shashua's tech empire," and the lows were real. Three of the companies that had long since left the lab and were expected to thrive, she wrote, had stumbled in the preceding year: OrCam had struggled financially, and One Zero had reported significant losses for 2023. Mobileye was the biggest drama of all. In 2026, a related Calcalist headline reported that Shashua would leave AI21 after nine years.
And in January 2026, the lab and the company folded into each other in a way that drew skeptical attention. Mobileye agreed to acquire Mentee Robotics, the humanoid company Shashua had co-founded four years earlier, for approximately $900 million, to continue operating as an independent unit. The strategic rationale was coherent: Mobileye was rebranding itself as a "physical AI" company, and a robot that walks through a kitchen is, in Shashua's framing, solving a cousin of the problem a car solves on a highway. "Robotics are a natural investment for autonomous tech companies," he told Automotive News at CES that month, adding that humanoids could help meet labor shortages. Calcalist's headline on the deal was less lyrical: "The question marks behind Mobileye's $900 million bet on Mentee." The chief executive of a public company was buying, with its shareholders' money, a startup he had co-founded. The deal may have been sound. It was, inescapably, also a mirror.
The Arithmetic of the Market
Physics forgives nothing, but it is at least consistent. Markets are another kind of road, full of aggressive players who make mistakes, and Shashua's formulas did not cover them.
By August 2024, Mobileye's stock had fallen 65 percent since the start of the year, dropping below the price Intel had paid in 2017 and below the price of its 2022 relisting. The company had lowered its forecasts repeatedly, the latest cut citing slowdown and volatility in the Chinese car market, which had become a major source of revenue through customers like Geely, whose Zeekr brand had launched with SuperVision's eleven cameras. A buildup of EyeQ chips in customers' inventories had to be worked off. Revenue for 2024 came in at $1.65 billion, down from $1.87 billion in 2022, and the company reported a net loss of $3.1 billion.
Paragon Intel, a firm that sells management assessments to hedge funds, published a case study in October 2025 whose headline was a verdict: "CEO Shashua's Tech Vision May Not Overcome Forecasting & Profitability Misses." It credited his "technical vision" for securing long-term design wins and named his "apparent blind spots in forecasting and profitability management." The questions it posed for investors had the shape of an exam the professor might set for a student. When confronted with data that contradicts his strategic narrative, does Shashua challenge the data or adapt?
In December 2025 Mobileye announced it would lay off about 200 employees, roughly 5 percent of its global workforce, most of them in Israel. The shares were down more than 40 percent for the year.
There is a temptation to read this as simple comeuppance, the visionary punished by the spreadsheet, and the temptation should be resisted, because the technical arguments kept going his way. Waymo's success with heavily redundant robotaxis looked like vindication of the multi-sensor philosophy he had adopted after 2016. Volkswagen applied his crowdsourced mapping technology, REM, to driver assistance globally; by January 2021, Mobileye said REM had mapped more than 7.5 billion kilometers of road, uploading only about 10 kilobytes of data per kilometer from ordinary cars. Porsche signed on for SuperVision in May 2023. Polestar 4 was slated for the eyes-off Chauffeur system. Brad Templeton, the veteran robocar analyst, called him in 2023 "one of the most astute players in the field," and then asked the question everyone was asking: "Can they deliver?"
The same year Shashua received the Israel Prize for lifetime achievement and exceptional contribution to the nation, 2023, Tel Aviv University's Board of Governors awarded him the Ramniceanu Prize in Economics. Within eighteen months, the market would be lecturing him on forecasting.
Not a Political Person
"I am not a political person."
That is how he began an op-ed in Yedioth Ahronoth's English edition in 2023, in the middle of the largest civil protests in Israel's history, against the Netanyahu government's plan to curtail the power of the judiciary. Executives of Israel's technology sector had become central to the opposition, and Shashua, the country's most decorated founder, had held back. He had believed "with all my heart," he wrote, that the prime minister, "whom I deeply respect," would ultimately do the right thing. He had held meetings and long discussions with relevant parties. "However, unfortunately, I am no longer convinced of that."
The essay is not eloquent in the way protest writing usually aims to be. It is procedural, almost legalistic, a governance argument made by someone who sits on boards. "The government received the mandate from the majority, but it does not mean it received the mandate to change the face of the nation." Fundamental changes require broad consensus. "We must seek consensus. If reaching an agreement is not possible, the process should be stopped. It must be stopped." And then the line that only a CEO would write: "In the business entrepreneurship world, a CEO of a company that destroys its values is held responsible, regardless of who or what caused the destruction of value."
He would have reason, a year later, to remember that sentence about himself.
It would be easy to frame the op-ed as an exception, a scientist stepping briefly into the arena. It is better understood as the visible part of a longer engagement with the question of who gets to belong to the Israeli economy he had helped build. As of 2024, he and his family had given away about $60 million, much of it through the Shashua Family Foundation, whose president is his wife, Anat Ramaty Shashua, who also chairs Orr Shalom, an organization serving children and youth at risk. The money flows to populations that the Startup Nation narrative tends to leave in the background: the ultra-Orthodox, Israeli Arabs, women in STEM, and the country's geographic periphery. During the pandemic, the family launched WE-19, a $35 million fund to support roughly 2,000 small businesses. Together the couple founded the Social Solidarity Foundation to help small businesses grow; their daughter, Inbar Shashua Bar-Nir, chairs it. He has funded scholarships at Tel Aviv University, where a kid from the ORT school in Givatayim once took his first degree.
A 2019 piece in Ami, an Orthodox Jewish magazine, carried a headline that captures the pattern better than any foundation brochure: "Developing Driverless Cars in Jerusalem: High-tech taps into chareidim to make the future a reality." Mobileye's headquarters are in Jerusalem, not in Tel Aviv's tower district or the Herzliya corridor. Jerusalem is the city of the Hebrew University, and it is also the city of Israel's divides, secular and religious, Jewish and Arab, rich and poor, laid on top of one another. The company that taught cars to navigate aggressive, uncooperative human traffic learned to do it, in its first camera-only autonomous demonstrations in January 2020, on Jerusalem's streets.
The Right Timing
The announcement came on July 23, 2026, alongside a forecast for a weaker third quarter, and it was framed as a matter of division of labor. "We are at a crossroads of very significant operational expansion," Shashua said. "We are entering the consumer market and a new market in robotics, all while the pace of development in AI is insane."
— Amnon Shashua, on the conference call announcing his resignation as Mobileye CEO, July 23, 2026I am originally a scientist and an entrepreneur. Doing both things at the same time, ongoing management and technological leadership, turned out to be too much of a burden.
He said he would focus on "the technological and strategic side and leave the operational expansion to new blood." He rejected the suggestion that his other ventures had distracted him. "Mobileye has always been my top priority. We currently have 4,200 employees who require close management care." He attributed the stock's decline to broader market dynamics. Calcalist ran the story under a headline quoting him: "The company is bigger than me." Calcalist's Meir Orbach, writing three days later, observed that the decision came against a backdrop of growing investor frustration with Mobileye's heavy research spending and the long timelines of autonomy, and that the founder's attention might now turn toward a newer AI venture, doubleAI.
It is tempting to read the resignation as the end of the Ullman formula, the day-a-week arrangement finally collapsing under the weight of 4,200 people. It is closer to a reversion. For most of three decades, the professor had been forced to be the manager too, because the monocular bet was his and no one else could defend it to an automaker's procurement committee or an Intel board. Now he was handing the hands to someone else and keeping the eye.
That spring, the U.S. National Academy of Engineering had elected him an international member for his contributions to computer vision and its application to autonomous driving. The year before, TIME had put him on its list of the hundred most influential people in artificial intelligence. Early in 2026, at the World Governments Summit, he had spent a fireside chat explaining what had changed in the decade since convolutional networks gave way to transformers, what could be learned from Waymo's early success, and why humanoid robots. The questions were still the questions of his master's thesis: what to see, what to ignore, how a machine decides what matters in a cluttered field.
The car is still the place where the answer gets tested. In tens of millions of vehicles on the road tonight, built by BMW and Ford and Volkswagen and Honda and Nissan and Geely and dozens of others whose drivers have never heard his name, a small lens sits against the inside of the windshield glass, looking ahead, comparing each frame to the last, every 27 milliseconds, a single eye mounted in the spot a young man once indicated to a friend whose name nobody wrote down, when he said: right here.
Part IIThe Playbook
What follows distills Shashua's career into operating principles. They are drawn from what he did, not what he said he did. In several cases, from what went wrong. His record is unusual among founders because it includes two IPOs, a record acquisition, a public rupture with his most famous customer, and a long, humbling stretch of forecasting misses, all under the same name. The lessons are correspondingly double-edged.
The principles
- 1.Subtract the expensive sensor.
- 2.Put the math in the silicon.
- 3.Never leave the lab.
- 4.Choose the hardest-to-win customer.
- 5.Formalize what everyone else leaves to judgment.
- 6.Give away the standard, sell the chip.
- 7.Fire the customer who misuses your product.
- 8.Refuse moonshots; build in blades.
- 9.Write your conviction into the contract.
- 10.Run one idea through many companies.
- 11.Treat the forecast as a technical problem.
- 12.Know which half of the job is yours.
Principle 1
Subtract the expensive sensor
The founding move of Mobileye was a subtraction. Every engineer in the room assumed depth perception needed two cameras, or radar, or both; Shashua's reply, "One is enough. When you shut one eye, you don't go blind," converted a perceptual fact into a cost structure. Monocular vision infers distance from change over time rather than disparity across space, which means the hardware can be cheap so long as the computation is clever. Cheap hardware meant driver assistance could leave the luxury segment, which meant volume, which meant tens of millions of units.
The firsts Shashua himself catalogued, camera-only forward-collision warning in 2011 and camera-only adaptive cruise in 2013, are each a sensor removed from the bill of materials. The pattern is not frugality. It is the recognition that in a mass market, every component you eliminate is a margin you own and a price point competitors cannot reach.
The second half of the lesson is in what came later. After 2016, for systems that let the driver look away, Mobileye added lidar and radar back as redundancy. Subtraction is a strategy for the mass product; it is not a theology.
Tactic: List every component in your product that the industry considers mandatory, and fund a small team to prove which one can be replaced by computation.
Principle 2
Put the math in the silicon
Mobileye could have been a software licensor. It chose to design the EyeQ system-on-chip, with its vision algorithms built into custom silicon, starting with a 180-nanometer, 2.5-watt part in 2008. That decision made the company far harder to displace. A carmaker could not swap Mobileye's software for a competitor's without redesigning the hardware, and the EPO's description of the system, with "sophisticated algorithms that are built into the hardware," captures why: the product was the fusion.
Owning the chip also let Mobileye set the cadence. Each EyeQ generation, roughly every three to four years, shrank the process node and lengthened the list of networks the chip could run, from vehicle rears and lane markings to traffic lights, road edges, animals, and trajectory prediction. Customers upgraded on Mobileye's schedule.
◉
The EyeQ Cadence
Each generation shrank the process and widened what a single camera could understand.
2008
EyeQ1 on 180 nm CMOS, ~0.0044 TOPS at 2.5 W. Ships in the BMW 7 Series via Continental; GM Cadillac lane departure warning.
2010
EyeQ2 on 90 nm. Pedestrian-detection warning arrives.
2014
EyeQ3 on 40 nm. Powers Tesla Autopilot Hardware 1 until the 2016 split.
2018
EyeQ4 on 28 nm FD-SOI. BMW, Nissan ProPilot, VW Travel Assist.
2021
EyeQ5 on 7 nm FinFET. Two EyeQ5H chips drive SuperVision in the Zeekr 001.
2025–27
EyeQ6 on 7 nm; EyeQ7 planned on 5 nm.
Tactic: If your advantage is an algorithm, find the physical artifact into which it can be embedded so that adopting a rival means rebuilding, not reinstalling.
Principle 3
Never leave the lab
Shashua copied Shimon Ullman's arrangement precisely: one day a week at the company, four at the university. He held his Hebrew University appointment through every stage of Mobileye's life, including a stint as head of the engineering school while the company was still pre-product. The effect was not merely sentimental. The university became Mobileye's recruiting pipeline and its research arm. Reuters counted at least 11 Mobileye directors and officers produced by its scientific community. The 2019 EPO team and the co-founder lists of Mentee and AAI read like a roster of his co-authors and former students.
The lab gave him something a pure operator lacks: a stream of people trained in his way of thinking, arriving every year, already fluent in the problems the company would face five years later. It also gave him intellectual legitimacy when he made unpopular technical arguments, against end-to-end networks, for formal safety models. Industry listens differently to a professor.
The cost is real, and he named it himself in 2026: doing technological leadership and ongoing management at once "turned out to be too much of a burden." The dual life is a moat. It is also a tax.
Tactic: Keep a formal, recurring commitment to the research community that produced your core idea, and treat it as your hiring strategy rather than a hobby.
Principle 4
Choose the hardest-to-win customer
Mobileye's first market was the most conservative buyer in manufacturing: global automakers and their Tier-1 suppliers, with multi-year design cycles, punishing qualification standards, and an existential fear of liability. German engineers told Shashua a camera-based system was impossible. Winning them took nearly a decade from founding to the first EyeQ1 in a BMW.
But once won, that customer is extraordinarily sticky. A design win lasts the life of a vehicle platform. By the end of 2014, Mobileye was in 160 models from 18 manufacturers; by 2020, in more than 55 million cars. CogniTens had taught Shashua, earlier, that the automotive supply chain would buy from a professor if the professor could prove his product in measurable terms. The difficulty of the sale was the barrier to entry.
Tactic: If your technology is defensible, sell first to the customer whose qualification process your competitors cannot survive, and let that process become your moat.
Principle 5
Formalize what everyone else leaves to judgment
Responsibility-Sensitive Safety is Shashua's most characteristic invention because it is not a product. It is a definition. Where the rest of the industry treated safe driving as a matter of accumulating miles and hoping the statistics converged, he and Shalev-Shwartz tried to write down what careful driving is: safe distances, cut-in responsibility, right of way, caution under occlusion. He backed it with data, six million accidents, 99.4 percent of which he said fell into 37 scenarios.
The power of a formal model is that it moves the argument. Once RSS existed, regulators, insurers, and competitors had to respond to it on its terms. The essay defending it shows the depth of the method: the refusal to let a car trade one crash for a "less severe" one, because the machine might miss "a baby in the back seat." Formalization, done honestly, includes formalizing what you cannot know.
Tactic: Identify the most consequential judgment call in your industry that no one has defined precisely, and publish the definition before anyone else does.
Principle 6
Give away the standard, sell the chip
Mobileye published the full mathematics of RSS, insisting that "the safety of automated vehicles should not be proprietary." Baidu and Valeo adopted it; a Chinese standards body formed a working group around it. That generosity was also strategy. A safety standard written by Mobileye's scientists naturally fits the architecture of Mobileye's systems. The more widely RSS spread, the more the industry's definition of "safe" resembled the thing Mobileye already sold.
The 2019 editorial against Nvidia shows the edge of this play. Open standards invite imitation, and imitation erodes credit. Shashua's response, to publicly list who announced what first, was a reminder that in an open-standards strategy, the authorship is the asset, and you must defend it loudly.
⇄
Conventional Wisdom vs. the Shashua Approach
Where he broke from the industry's defaults, and what each break bought.
| Conventional wisdom | Shashua's approach | Outcome |
|---|---|---|
| Depth needs two cameras or radar. | One camera, motion over time, computed every 27 ms. | Scale ADAS leaves the luxury segment. |
| License the software. | Embed it in custom EyeQ silicon. | Scale Platform lock-in across generations. |
| Safety is proven by miles driven. | Safety is defined mathematically (RSS). | Standard Adopted by Baidu, Valeo; China ITS working group. |
| Keep your safety IP proprietary. | Publish it openly. | Trade-off Influence gained; imitation follows. |
| Never walk away from a marquee customer. | End the Tesla relationship after the 2016 crash. | Trade-off Brand on safety protected; halo lost. |
| End-to-end neural nets will solve driving. | Embed learning inside engineered architecture. | Open Contested; the bet remains live. |
| Founders cash out at acquisition. | Waive acceleration; vest over four years; report only to Intel's CEO. | Scale Autonomy preserved inside Intel. |
Tactic: If you need the industry to converge on a standard, write it yourself, publish it free, and make sure your product is the most natural implementation of it.
Principle 7
Fire the customer who misuses your product
Tesla was Mobileye's most visible customer, and Shashua's own biography had boasted of the "autopilot" feature it enabled. After the May 2016 crash, Mobileye ended the relationship and Shashua said publicly that Tesla "was pushing the envelope in terms of safety," insisting Autopilot was "a driver assistance system and not a driverless system."
The lesson is not that you should abandon customers in a crisis. It is that a component supplier's brand is only as safe as the most aggressive way its customer markets the component. Mobileye's system had worked as designed and still faced reputational exposure because its customer's product name invited drivers to stop paying attention. For a supplier whose entire value proposition was safety, the halo was not worth the risk. The cost was real, and Shashua accepted it.
Tactic: Write down in advance the uses of your product you will not support, and be prepared to end your most prestigious relationship if a customer crosses that line.
Principle 8
Refuse moonshots; build in blades
Shashua's January 2023 explanation of Mobileye's roadmap is a compact theory of R&D risk. Without an "incremental and modular transition," he wrote, "development becomes a series of 'moonshots' where the development and validation of one blade is discarded and a completely new development and validation is required for the next." So Mobileye concentrated its hardest work in SuperVision, a camera-only, hands-off, eyes-on system, and treated the higher levels of autonomy as additions of redundant sensors rather than new programs.
The approach lets each generation fund the next. ADAS revenue paid for SuperVision; SuperVision's perception stack becomes Chauffeur's and Drive's foundation. It is the opposite of the venture-backed robotaxi model, which burns capital for years on a single leap. It is also why the market's frustration, about long timelines and heavy R&D, never quite matched the engineering logic. Incremental is slower to dazzle and harder to kill.
Tactic: Structure your roadmap so that every product generation is a strict subset of the next, and refuse projects whose validation work will be discarded.
Principle 9
Write your conviction into the contract
The March 12, 2017 addendum to Shashua's employment agreement is a masterclass in founder positioning. He waived the acceleration of his unvested equity, deferring all of it for three years and half of it for four, signaling to Intel that he was staying. In exchange, he secured a direct reporting line to Intel's chief executive, with a clause that any change to that line would count as "Deemed Dismissal" and vest everything immediately.
The two clauses buy each other. The first gives the acquirer confidence. The second makes it prohibitively expensive for the acquirer to bury the founder in its hierarchy. Mobileye remained in Jerusalem, remained intact, and five years later was spun back out at a higher valuation than Intel paid.
Tactic: When selling your company, trade liquidity for structural guarantees, and price the guarantees by attaching your equity to them.
Principle 10
Run one idea through many companies
OrCam, AI21, One Zero, Mentee, AAI: on the surface, a scattered portfolio across wearables, language models, banking, humanoids, and research. Underneath, one thesis, applied AI from a Jerusalem research group, run through different markets. OrCam is Mobileye's camera turned toward the page. Mentee is Mobileye's perception and planning turned toward a kitchen. The co-founders are recurring co-authors.
The portfolio has had mixed results, and the record should be read whole.
◇
The Ventures Lab
Six companies, one research group, uneven outcomes.
| Company | Founded | Domain | Status in the record |
|---|---|---|---|
| CogniTens | 1995 | 3D optical inspection | Exit Sold to Hexagon AB, 2007 |
| Mobileye | 1999 | ADAS, autonomy, physical AI | Exit $15.3B to Intel; relisted 2022 |
| OrCam | 2010 | Assistive vision | Strained Financial struggles reported, 2024 |
| AI21 Labs | 2017 | Language models | Transition $155M Series C, 2023; Shashua departing after nine years |
| One Zero | ~2020 (licensed 2022) | Digital banking | Strained Significant 2023 losses |
| Mentee Robotics | 2022 | Humanoid robots | Exit Acquired by Mobileye, ~$900M, 2026 |
| AAI Technologies | 2023–24 | Expert AI | Scaling Raised at $1B+ valuation, 2025 |
The Mentee sale shows the risk of the model. When the ventures lab and the public company overlap, every intra-family transaction invites a governance question, which Calcalist asked in its headline: "The question marks behind Mobileye's $900 million bet on Mentee." A portfolio built around one founder's thesis is coherent. It is also, from outside, hard to distinguish from a founder dealing with himself.
Tactic: Spin out ventures that share your core technical thesis, but build the governance walls between them before you need them, not after an intercompany deal forces the question.
Principle 11
Treat the forecast as a technical problem
This is the cautionary principle. Shashua's technical calls have aged well: his skepticism of end-to-end systems, his embrace of redundancy, his early bet on crowdsourced mapping. His financial calls have not. In 2024, Mobileye repeatedly lowered guidance, worked through an EyeQ inventory glut, and lost 65 percent of its market value in seven months. Paragon Intel's assessment named "apparent blind spots in forecasting and profitability management." In December 2025 came 200 layoffs.
The irony is precise. Shashua built his reputation by refusing to let safety rest on intuition, by insisting on formal models where others relied on judgment. Demand forecasting, especially across volatile Chinese automakers, deserved the same rigor and visibly did not receive it at the level the market expected. The same mind that reduced six million accidents to 37 scenarios did not reduce its own order book to anything investors trusted.
Tactic: Apply your most rigorous internal discipline to the function you personally find least interesting, because that is where your blind spot will be priced.
Principle 12
Know which half of the job is yours
Shashua's resignation statement is unusually candid for a founder: "I am originally a scientist and an entrepreneur. Doing both things at the same time, ongoing management and technological leadership, turned out to be too much of a burden." He chose "the technological and strategic side" and left "the operational expansion to new blood."
The Ullman formula of a day a week at the company had always presumed someone else did the other four. In Mobileye's early years that was Ziv Aviram. The arrangement worked best when the division was explicit. The resignation is the formula reasserted at a scale of 4,200 employees: the founder recognizing, late but not too late, that his comparative advantage is the eye, not the hands.
Tactic: Decide which half of the CEO job you are irreplaceable at, hire a peer for the other half while you still have the leverage to choose, and say publicly why.
Free playbook
Get The Business Model Playbook
58 business models, one visual page each: how the money flows, the metrics that matter, and who runs it. Free when you join the Faster Than Normal email.
Free. No spam. Unsubscribe anytime.
Part IIIQuotes and Maxims
In their words
— Amnon Shashua, Weizmann Institute interview, 2014Without the knowledge gained in Shimon Ullman's lab and watching him work the way he did — always thinking about how his lab insights might help make life better for people — I wouldn't be where I am today.
— Amnon Shashua, Intel editorial, March 25, 2019We openly shared all the technical details and mathematics behind RSS because we believe that the safety of automated vehicles should not be proprietary.
— Amnon Shashua, on end-to-end self-driving, LinkedIn, 2023Even the most advanced LLMs make embarrassing mistakes. Will we trust them to make safety-critical decisions?
— Amnon Shashua, on reasoning in large language models, LinkedIn, 2023Humans can find "rules" (algorithms, theories) to explain observations (data). Machines so far can memorize big data, but if AI would ever become a "great thinker/scientist" it needs to be able to "abstract" from it like humans, and not only memorize it.
— Amnon Shashua, op-ed on Israel's judicial overhaul, Ynetnews, 2023In the business entrepreneurship world, a CEO of a company that destroys its values is held responsible, regardless of who or what caused the destruction of value.
Maxims
- One eye is enough. The sensor everyone calls mandatory is often a computation nobody has bothered to write.
- Cheap glass, expensive math. Put your cost in what you can copy for free and your savings in what the customer has to buy.
- The lab is the moat. A founder who never leaves the university never stops recruiting from it.
- Define the word before you sell the product. Whoever formalizes "safe" sets the terms of every argument that follows.
- Give away the standard. Influence compounds faster than licensing fees, so long as you defend the authorship.
- Your customer's marketing is your liability. A supplier's reputation ends where its most reckless client's product name begins.
- No moonshots. Build each generation so the next one inherits its work instead of discarding it.
- Price your conviction in equity. Defer your own payout to buy the structure that protects the company.
- Formalize what you can't know. The baby in the back seat is the variable that breaks every confident trade-off.
- Choose your half. The scientist who built the eye need not also be the hands.
