Part IThe Story
In 2011, at the airport in Amman, a young Jordanian programmer stood at a counter holding a visa the officials in front of him did not recognize. It was an O-1, the American visa reserved for people of "extraordinary ability." It is the classification used for Olympic athletes, Nobel-adjacent scientists, film directors with festival laurels. The Jordanian officials had apparently never seen one issued to a twenty-three-year-old whose credentials were a computer science degree from Princess Sumaya University for Technology and an open-source code editor that had briefly caught fire on Hacker News. They were not sure he would be allowed to stay. So, before they would let him board, they made him buy a return ticket home.
It consumed most of his cash. Amjad Masad arrived in the United States, by the account later given to Endeavor's chroniclers, with roughly $100 in his pocket, and, by another account, with nothing much besides credit card debt. Fifteen years later Forbes would estimate his net worth at $2 billion. The company he co-founded with his wife and his brother, Replit, would be valued at $9 billion and tracking, he said, toward a billion dollars in annual run-rate revenue. In 2024 it had made $2.8 million.
Those are the bookends that get repeated, and they make the familiar shape: immigrant arrives with nothing, becomes a billionaire, America works. Masad has told a version of that story himself, in a widely circulated Twitter thread listing ten things he loved about the country, which Peggy Noonan picked up in The Wall Street Journal. It is not a false story. It also leaves out nearly everything interesting. It leaves out the return ticket, for one, the bureaucratic insistence that he was the sort of person who comes back. It leaves out the eight years in which Replit, by his own description, "wasn't successful commercially," and the morning in early 2024 when he bet the company on a demo that malfunctioned and began answering in Chinese. It leaves out the autumn after October 7, 2023, when party invitations dried up and investors in his own cap table called him a "terrorist sympathizer." It leaves out the part where the same man who spent fifteen years telling the world to learn to code began telling it not to bother.
Replit takes its name from the REPL, the read–evaluate–print loop, which is the simplest possible conversation between a person and a computer. You type a line. The machine reads it, evaluates it, prints the answer, and waits for the next line. Then it loops. The return ticket, it turns out, was the first artifact of a life that keeps closing its loops: the father's computer and the son's company, the hacked university and its honorary doctorate, the boy who wanted to make software and the agent that now makes it for anyone who asks.
By the Numbers
The Replit Arc
$9BValuation at the $400M Series D, March 2026 — triple its value six months earlier
$2.8MReplit's revenue in all of 2024, before the Agent took hold
4Y Combinator rejections before the company got in, by most tellings
$1BAcquisition offer from a big tech company, declined in 2021
130 → 65Headcount before and after the 2024 bet on AI agents
300%Net revenue retention at the high end, per Masad in May 2026
50M+People Replit says have built on its platform
Everything Downstream of a Loan
The first computer in the neighborhood arrived in 1993. Masad was six. His father, a civil engineer working for the Jordanian government, did not have the money for it and took out a loan. Elsewhere Masad has said his father plowed his life savings into the machine; the two accounts are not incompatible, and both describe the same posture: a man of modest salary staking something real on an object whose purpose he could not quite articulate. "If you'd asked him why it mattered, I don't think he could have told you," Masad wrote decades later. It was an IBM PC running MS-DOS. To the boy it looked like "alien technology." In another telling, more vivid: "What is this alien species doing here?"
The family had some experience with arrival. Masad's father was Palestinian, born in Syria to a refugee family and later moved to Amman, back when, as Masad put it, Amman was only beginning to become a major city in the region. His mother's people were Algerian; they had fled the war against the French, gone to Syria, and then gone on again to Amman. "I come from, I guess, multiple generation of immigrants," he told the Atlas Society in 2022, a sentence that compresses something like seventy years of the twentieth-century Arab world into a shrug. In his own family's telling, three nations' worth of displacement eventually came to rest in a government engineer's apartment in Amman, where there was now a computer nobody could afford.
The father struggled with it. One of Masad's earliest memories is of standing behind him while he worked through an enormous manual, finger-typing commands. CD. MKDIR. (In a 2026 LinkedIn post Masad listed
LS among them, a Unix command that does not exist in DOS, and a commenter politely noted the error. Memory is a lossy compression format.) After his father went to bed, the boy would sneak back in and tinker. The machine did exactly what it was told. That was the hook: not the games, though there were games, but the obedience.Then one day his father walked in and found the computer in pieces on the floor. The loan, the life savings, the alien technology, disassembled. He was furious, until his son told him: wait, I know how to put it back together.
"That's how I learned hardware," Masad wrote, "and the software came right after."
The software came in a box of CDs at a computer show in Jordan, an annual event where people displayed the latest machines. Among the accounting packages and other boring things was a visual programming environment, logic blocks that fired on events. He understood it immediately. The first thing he built was a game to teach math to his younger brother: the screen would show a partial equation, X plus 10 equals 19, and if the brother guessed nine, the computer played the sound of applause. Decades later, describing this to an interviewer in 2020, Masad added a parenthetical: the brother, he said, now worked with him at Repl.it. The first user was family. The first product was a lesson. The first interface was a machine that clapped.
By his early teens he had a real programming environment, Visual Basic, and an obsession that was not programming at all. It was Counter-Strike. "Whatever money, scraps of money I have, I would, like, put it into that and just play," he recalled. He played in the LAN gaming cafés of Amman, places where dial-up was too slow and so the internet came in on cables strung between rented machines. He got good enough that it became a source of income, winning tournaments in what he would later recognize as early esports. And he noticed how the cafés were run. You wrote your name in a notebook. You played. Then someone tapped you on the shoulder and told you your time was up.
"This is ridiculous," he told the owners. "You should be using software."
What followed has the cadence of a founding myth, which is perhaps why he tells it so economically: "I went into a hole and a year and a half later emerged." He came out with client-server management software for the cafés, written in VB4, and went back to the places where he used to play and sold them licenses. Five hundred dollars a year.
It is worth pausing on what was already present in that teenager. He found friction in a system he used every day. He was offended by it, a little theatrically. He built the replacement himself rather than waiting for someone else to. And he sold it back to the people who had been tapping him on the shoulder. The notebook in the café was the first committee he ever killed.
Banned and Passed
At Princess Sumaya University for Technology, a computer science program in Amman founded under the umbrella of the Royal Scientific Society, Masad aced his coursework and could not make himself attend class. The university's rules were not interested in his grades. Miss too many lectures and you failed automatically. He spent five years in a four-year program.
In 2008 he hacked into the university's system to change his grades. "I just wanted to see if I could do it and how it worked," he told Joe Rogan years later, which is the kind of explanation that is simultaneously completely sincere and completely inadequate. What began as private rebellion became a public incident: the intrusion crashed the registration database. The anomaly that tripped the alarms was a single student record that the system flagged as simultaneously banned and passed. The head of registration called the Masad family landline.
He graduated in 2010.
The episode sits awkwardly in the canonical founder biography, the way such episodes do. Tech culture loves a hacker origin story right up until the hacking involves an actual institution's actual records, at which point everyone prefers to call it curiosity. Masad himself seems unbothered by the ambiguity. But the image is too precise to waste: a computer system forced to hold two contradictory states about the same person, because a person had refused to accept the system's description of him. Banned and passed. It would not be the last time an institution could not decide which one he was.
In parallel he was working, and in a register that would matter. He did Windows system administration, IT consulting, web application security. The security work paid and taught him how systems break, but it bored him in a specific way he could later name with precision: "When you're doing things like security auditing, you're just writing code for the immediate task, but you're not really thinking like a software engineer."
So he went looking for a post-graduate education he could administer himself. He researched what MIT taught its first-year computer science students and found Structure and Interpretation of Computer Programs, by Harold Abelson and Gerald and Julie Sussman, which teaches through a dialect of Lisp called Scheme. He didn't have a laptop. Scheme was an old language, poorly supported online. Here, in miniature, was the whole Replit problem: a curious person, a canonical text, and no environment in which to actually run the code. The friction was not intellectual. It was infrastructural.
Scheme led him to The Little Schemer, Daniel Friedman's strange, catechism-shaped book, and to the first time he derived the Y combinator, the recursive function that lets a function call itself without having a name. He described the feeling with a borrowed phrase: "the intoxicating delight of sudden understanding." Lisp led him to Paul Graham's essays, Graham having made his fortune on a Lisp-built company and his reputation writing about it. "Which was, of course," Masad wrote, "a gateway drug to startups." Then Jessica Livingston's Founders at Work, which he credits with demystifying startups and humanizing founders, with making it seem that a Silicon Valley company was something a person might actually start. Then Hacker News, which he read daily for years.
There is a pleasing recursion here that the author of a Y combinator derivation would appreciate. A Jordanian student learns a fixed-point combinator from a book. The book leads him to the essays of the man who named his accelerator after that same combinator. The accelerator will reject him repeatedly. Its founder will eventually be the one who brings him in.
A Fun Toy
In 2009, with browsers improving fast and Google Docs teaching a generation that a document could live in a URL, Masad had the thought that would occupy the next fifteen years of his life. "Man, I can just share these things easily," he remembered thinking. "Why can't I share code? I mean, that's what the browser is." He had tried to write implementations of other programming languages in JavaScript and found it impractical. But the browser was getting fast enough that it might be possible to compile languages into JavaScript and run them there, no installation, no dependency hell, no "works on my machine."
The result, built with Haya Odeh over roughly two years and released as open source in 2011, was called JSRepl. When he posted it to Hacker News it went viral. Brendan Eich, the creator of JavaScript, took notice. So did two companies building the new online-education industry. JSRepl ended up powering the in-browser tutorials of both Udacity and Codecademy, and a journalist writing for a Japanese engineering magazine in 2012 reduced Masad's career to a perfect two-clause sentence: "Build an IDE, get a job in New York."
Haya Odeh's path rhymes with his at an angle. She was born in Abu Dhabi and raised in Jordan, studied graphic design and fine arts at Al-Ahliyya Amman University, worked at design studios in Amman, and, after moving to the United States, freelanced for clients including the Clinton Foundation and the consultancy Organic. Where Masad came to computing through the machine's obedience, she came through the psychology of visual communication, through the question of what makes an interface feel like something a person is allowed to touch. She would lead design at Replit as a vice president, and her name would sit beside his when Endeavor Jordan selected them in 2026 as its first unicorn founders from the country.
At Codecademy, which had just graduated from Y Combinator's Summer 2011 batch, Masad became employee number one, the lead front-end engineer, building product and infrastructure from November 2011 to October 2013. "Being that early at a high-growth consumer startup is a great, albeit compressed, learning experience," he wrote later, in the careful voice of a man describing something exhausting. He had come to America on a visa for extraordinary ability, and he was spending his days teaching strangers how to write
for loops.New York also gave him something less expected: improv. Masad's first public talk had sent his heart rate to 150 beats per minute. So he took improv and storytelling classes, a course of what he calls, without irony, desensitization. "The best way to get desensitized is to just make a fool of yourself every day doing improv." It is easy to skip past this detail, and a mistake. A decade later he would say the company would have died without his public voice, and that voice was built deliberately, rep by rep, in rooms full of amateurs in New York.
In October 2013 he moved to Facebook, where he helped start the JavaScript infrastructure team and became its tech lead. The team built and maintained the open-source plumbing of the modern web: the Babel compiler, the Jest testing framework, the React Native packager. His GitHub profile still pins those repositories, collectively starred hundreds of thousands of times, a fossil record of the years when Masad made other programmers' tools better. He once called the Facebook red book, the company's internal culture manifesto, "the best startup culture book every written." He was at Facebook until April 2016. And then, because no one else had built it, he quit to build the thing he had imagined in 2009.
"I've been waiting for someone to start this startup for years," he wrote. In 2016, "nobody seemed to be coming close."
Before quitting, he applied to Y Combinator's open office hours to see if the idea was the kind of thing they would back. The partner he met told him that an online REPL was "not really a startup — it's just a fun toy," and suggested that he go join another company building a superficially similar technology instead.
He and Odeh quit their jobs anyway. They allocated $20,000 to the venture and worked out of their home in Foster City, a planned community on landfill along the edge of San Francisco Bay. (Max Shawabkeh, once listed as a co-founder, left early on.) Replit monetized faster than most toys do, selling its code-evaluation infrastructure as an API. But the free service grew faster than the revenue, and Masad did not believe he could raise money; his startup sat at the intersection of developer tools and education technology, two categories he described, accurately, as having "comparatively yielded minimal returns to VCs."
Roy Bahat, who runs Bloomberg Beta and whom Masad had met during the Codecademy years, decided to back them anyway. With that stamp, they raised a small seed round. Around the same time, Y Combinator rejected their application. It would take, by most tellings, four rejections before Paul Graham read about the project on Hacker News and encouraged them to apply again. Replit joined the Winter 2018 batch.
The stretch in between is the least glamorous part of the story and possibly the most revealing. Following Graham's dictum that a startup should be "default alive," Replit built a commercial product for computer science teachers to manage their classrooms. Masad learned the American school sales process the way a pilgrim learns a mountain: "purchase-orders, lost-in-the-mail checks, and poor teachers paying out-of-pocket to avoid the bureaucracy." The vision, he wrote at the time, was moral as much as commercial: "Increasing access is a moral, economic, and political imperative." The business was lost checks.
Teams for Education eventually stopped receiving bug fixes and features on November 15, 2023. By then the company had other things on its mind.
Memeing a Dream
For most of a decade, Replit was a company whose story outran its income statement, and Masad knew it.
— Amjad Masad, on the a16z podcast, 2026Replit would have probably died if I wasn't telling a story that is larger than the company itself.
"The company wasn't successful commercially," he told Andreessen Horowitz's Erik Torenberg, "but we've been very successful at fundraising, recruiting." The phrase he used for the work was strange and exact: "if you're trying to meme a dream into reality... you have to play that game." Then, with the vindicated man's flourish that also contains a confession: "It turns out everything I said was true, but at the time it wasn't real. It was a dream."
The dream was large. On his personal website, Masad wrote that computers are the most powerful tools in human history, and that most people are "mere consumers of these machines." He reached for the printing press, which took about a century to turn from invention into democratic, scientific, and industrial revolution, and argued that software was at the same stage, held by a priesthood. "It's creating an unbalanced world where there are programmers, and then there are those who are programmed." Replit would "bring the next billion software creators online." And then, more audaciously: "If we're successful, anyone who's willing to learn and generate good ideas will be wealthy."
At a 2023 gathering hosted by the venture firm 8VC he offered a darker version he called "the Steve Jobs Black Pill." Apple, in making computers beautiful and closed, had created a schism between makers and consumers. You once had to learn to program a computer to use it; now "most people use computers as televisions." Masad loved Jobs, everyone in the Valley does, and still accused him, in effect, of turning citizens into audiences. It is a curious critique from the founder of a consumer software company, and he would spend the next three years trying to resolve it by making the audience into authors.
"We were talking about creating a billion software creators back in 2018 at YC," he recalled in 2026, "and people sometimes actually laughed at that dream."
The laughing stopped first for Paul Graham. On December 2, 2020, Graham tweeted a chart of Replit's user base: "This graph of Replit's user base is very impressive, not just for its perfect exponential shape, but for the numbers. 5 million users is a lot when those users are programmers." In February 2021 the company raised $20 million. By the time First Round Capital interviewed him, Replit claimed more than 30 million users, had raised some $200 million in venture funding, and carried a $1.2 billion valuation. Users and valuation, though, are not revenue. "For years," Masad wrote, "our entire marketing department was my Twitter account."
Then, in 2021, a big technology company offered to buy Replit for a billion dollars.
"We were six people, and broke," he wrote years later. (The headcount figure is his; the company had raised $20 million that year, so "broke" presumably described the gap between users and income rather than the bank balance, and one skeptical commenter immediately did the arithmetic on what multiple a billion dollars would have been on roughly a million in revenue.) "We had plenty of users but little revenue to show. It was a tough time both at the company and personally. My mother was fighting cancer back in Jordan."
He thought about it for a long time and said no. Part of it was that big companies acquire startups and then stop developing them. The bigger part was a question he could not get past: if I sell, what do I do next? He had watched founders sell. Most never built anything meaningful again. "A lot of them get depressed." And he already knew what he would want to work on, which was this. He would end up building the same thing again from zero, without the team or the users. "Selling wouldn't have ended the work. It would have just ended the progress."
Years later, in what he describes as "a much worse stretch," Graham asked him whether Replit was a phase or the thing he wanted to keep doing. He could not picture doing anything else. Graham's reply, as Masad recounts it:
"Then why does a bad month matter? Or a bad year? You have decades."
It is worth noticing that the man who once told Masad the wrong thing (Graham had also, Masad has said, given specific and confidently incorrect advice about implementing BASIC on Replit) was the same man who told him the one thing that mattered. Advice, Masad came to believe, is time- and context-dependent, and people overgeneralize from their own experience. Some sentences, though, travel.
The Demo That Answered in Chinese
In 2015, Masad put a line into his seed deck describing a product he had no idea how to build. The line was about blurring the distinction between learning and building: you shouldn't have to learn to code before you can make something; you should learn by making it. "It was a nice slide but the tools didn't exist. So we kept picking it up and putting it back down."
The 2016 version of the master plan, as he later recounted it to Reach Capital, had three bullets. Grow the population of young coders through educators and schools. Build a collaborative environment assisted by AI, using the network's data to "build tools that no one else could build." Get to a point where you could not only learn on the platform but deploy and build a business entirely. Write your first line of code, earn your first dollar, build your first startup. It reads, in retrospect, like a prophecy. At the time it read like a pitch deck.
He had a theoretical basis for the second bullet before almost anyone. Years earlier he had read a paper, On the Naturalness of Software, arguing that code behaves statistically like natural language. So in 2019, when he ran OpenAI's GPT-2 on Replit and played with it, a model generating text "didn't feel far from a model generating code." Still nothing he could ship.
The first AI feature, when it came, was modest: highlight code, right-click, explain it. Then an assistant that helped you write code faster. "It was good," he wrote. "But it only helped people who already knew how to code. That was the part we were trying to fix."
In 2021 he went around the Valley with a pitch he could not sell. Replit had millions of users and more coding data than almost anyone. Train a coding model with us. "OpenAI. Google. Anthropic. Nobody bit. It wasn't that they thought I was wrong. It just wasn't a priority. The real race was language — chatbots, search, assistants. Coding was a side quest."
One small research team at Google understood. The deal came close and fell through. The researcher who had run that team was Michele Catasta, and when the deal died, Masad persuaded him to join Replit part-time. The pitch was absurd on its face: a company doing roughly a million dollars in revenue would train its own state-of-the-art coding model. "A radical thing to do in 2023," Masad wrote, with the satisfaction of someone writing after the fact. They shipped replit-code-3b. Catasta is now Replit's President and Head of AI. In the Masad cosmology, talent is something you discover in a niche before the market prices it, and a researcher whose own employer considered coding a side quest is exactly the kind of person you go get.
Agents were harder. Replit spent a couple of years on agent research that did not work. The models had almost no memory. "An agent couldn't hold a thought long enough to finish anything."
And then, early in 2024, an engineer showed him a demo. The model wrote code, ran it, debugged it. Then it went off the rails and started answering in Chinese.
"A broken demo," Masad wrote. "But I'd just watched a machine do the whole job for a full minute. We bet the company on it."
There is a particular kind of founder judgment on display here, and it is not the kind usually celebrated. Anyone can recognize a working product. What Masad claims to have seen in that minute was a derivative, not a value: not where the model was, but which way it was moving, and how fast. The Chinese was noise. The minute was signal.
The bet had a cost, and he has been more candid about it than most founders. "There was a bit of a founder mode story there," he told Semafor's Reed Albergotti in January 2025, "where the company has gotten big, the culture has gotten a little slow, a little soft." He created what he called the Agent Task Force, the best people in the company, and told them the mission had changed. "Replit is no longer about coding. Replit is about making software." It meant a layoff. It meant telling people who had joined a coding company that it was not one anymore. Those outside the task force felt, he said, that they weren't working on something important, and many left. "We lost half of the team. We were 130, now we're about 65."
The ones who stayed shipped Replit Agent in September 2024. In his own count: "Nine years after the slide."
What happened next is the part of the story that gets told as if it were the whole story. Within five or six months, revenue was up five times. Revenue for all of 2024 had been $2.8 million; within roughly a year the company's annualized revenue had passed $150 million, and Masad would later describe the run as going from $2.5 million to $250 million "in just over a year." In September 2025 Replit raised $250 million from Prysm, Andreessen Horowitz, Amex Ventures, and others at a $3 billion valuation. In March 2026 it raised $400 million more, led by Georgian with participation from the Qatar Investment Authority, Andreessen Horowitz, Coatue, Craft Ventures, and Y Combinator, at $9 billion. The accelerator that had rejected him four times was now buying in at a triple.
"Now every lab has a coding model," he wrote. "Coding is the benchmark everyone competes on. The side quest became the main event. Everyone got code-pilled. We just got there a few years early."
Those Who Are Programmed
The cruelest thing about getting what you want is that it changes what you wanted.
For fifteen years Masad's case had been that coding was literacy, and literacy was power: the printing press, the priesthood, the programmers and the programmed. Replit was a place to learn. Then the machine learned faster than the students, and in 2025 Masad started telling people they didn't need to learn to code at all. It turned out, he admitted to Turing Post, "to be a very controversial statement." His defense was that the mission had never been the syntax. "Code is a means to program — it's not programming itself." Punch cards gave way to assembly, assembly to C, C to JavaScript. The next abstraction was agents. He had expected it to take five or ten years. He now thought two.
"As an entrepreneur, you have to be honest with yourself," he told Platformer's Casey Newton. "Otherwise the company would die."
It is possible to read this as a betrayal of the original promise, and some of his early users do. Under a 2025 LinkedIn post, a woman who had first come to Replit as a beginner at a hackathon wrote that the company had once been "a community of young, creative people who learned and grew together," and that it now felt "more like a product than a shared mission." It is also possible to read the turn as the promise's fulfillment. The original sentence on his website said nothing about syntax. It said "anyone who's willing to learn and generate good ideas." The learning part had simply moved.
His favorite evidence for the second reading is concrete. At the TED AI conference in October 2023, he told the story of Akashdeep, a young man in India whose only computer was an Android phone, who learned to code on Replit and began earning more than his family's means. He told the story of Yoeri, a technical recruiter, not a software engineer, who built a startup on the platform. The two of them ended up working together. "AI is not just a tool," Masad said, "it's a ladder that equalizes opportunity." By 2023 Replit was running a small bounty economy in which people outsourced coding projects to each other; some were making as much as $10,000 a month, and the company was paying creators not only in the United States and India but increasingly in Nigeria, Kenya, Sri Lanka, and Malaysia.
The canonical enterprise parable is a non-technical program manager on Zillow's agent performance team who had ideas for the company's lead-routing system that never made it onto the engineering roadmap. He built and deployed a new routing algorithm himself on Replit. Masad says it produced tens of millions of dollars in incremental revenue. In one telling the man has since been promoted several times and now advises on AI strategy at the board level. "Customers aren't buying code generation," Masad told G Squared; "they're buying outcomes." It is the café notebook all over again, now inside a public company, and the person writing the replacement software is no longer the programmer. It is the man who was being tapped on the shoulder.
There is something deeply Masad about the fact that his son is his favorite demonstration. One Saturday morning, according to a HubSpot profile, his five-year-old turned to him with an idea: a game where the characters turn into frogs if they touch lava. They opened a browser. The child described the rules. A few prompts later, they were playing it. The father who once built a math game for his younger brother, with clapping for correct answers, now watches his son build games by describing them. Two generations, one feature request.
The same man reads deeply in philosophy of mind, the subject that, he says, kept him interested in computers as a child. Douglas Hofstadter's I Am a Strange Loop is among the books he cites. He lifts heavy things, likes grilling steaks, professes an interest in tradition and paleoanthropology, and argues that "the past holds more wisdom than people like to admit." His favorite films are old science fiction: Forbidden Planet, Brazil, Colossus: The Forbin Project, a 1970 film in which a defense supercomputer takes control of humanity for its own good. His favorite science fiction book is The City and the Stars, set in the last city on Earth, where machines run everything and nobody dies. In the version of the story he likes to tell, the plot turns on a young man born without the inherited memories everyone else carries, who asks what lies beyond the city walls. Masad uses it to argue that the social function of young people is to question what everyone else considers settled.
A man who builds machines that write software for people keeps returning to stories about machines that run people. He does not seem troubled by the tension, or perhaps he is the only one paying close enough attention to it.
Expelled
"Should I wear a keffiyeh to the shooting range?"
That was the question Masad put to Margaux MacColl of The San Francisco Standard in November 2025 as she slid into the passenger seat of his black Mercedes sports car. He wrapped the patterned Palestinian scarf around his neck anyway. At the range in Santa Clara he took an AR-22 and fired at a brisk clip into a target printed with a cartoon burglar. In less than two minutes the burglar's head was perfectly pocked with holes. MacColl suggested he should compete.
He smirked. "I always compete."
He was thirty-eight. Since October 7, 2023, when Hamas attacked Israel and the war in Gaza began, Masad had felt obliged to speak, and he did so publicly and repeatedly, calling out people in tech who, in his view, had supported Israel's "genocide" of the Palestinian people. He learned quickly how unpopular that was in Silicon Valley. Party invitations dried up. Group chats lit up with condemnation of his posts. Investors called him a "terrorist sympathizer." In July 2025, a member of a firm that had backed Replit publicly declared that he would donate any earnings from the investment to the Israel Defense Forces.
— Amjad Masad, to The San Francisco Standard, November 2025I felt like I was sort of expelled from Silicon Valley.
The timing is the kind of thing a novelist would be told to revise. Masad's public isolation coincided almost exactly with the company's resurrection. The demo that answered in Chinese arrived a few months into the backlash. The Agent shipped a year into it. The $3 billion valuation landed two months after an investor pledged his gains to the IDF. The industry that froze him out could not, it turned out, stop buying his product.
He did not modulate. Responding to people who told him his words had been hurtful, he posted on X a line that MacColl's editors used as the epigraph for her profile: "I must apologize to — absolutely nobody." On the San Francisco Standard's podcast in February 2026 he went further, and with something like relish: "Contrarianism in Silicon Valley is largely fake. Taking the Palestinian position on that issue was actually contrarian." It is a provocation aimed precisely at an industry that worships the word contrarian, the founder mythology of Peter Thiel's secrets and Paul Graham's unpopular ideas, and its effect is to accuse his peers of cosplaying the trait he paid for.
His theory of survival, expounded later to Andreessen Horowitz's audience, was severe: "Being canceled is a choice. You can choose to get canceled and retreat from the public eye, but I think if you're still out there, at some point, honestly, the haters kind of give up." It is a striking thing to say on the podcast of a firm whose founders had loudly supported Israel, and that also led his last two rounds. Silicon Valley, for all its tribalism, remained a market, and markets have short memories when the revenue chart goes vertical.
The keffiyeh is the detail that stays. Not the political argument but the garment, worn deliberately to a gun range in Santa Clara, by a man who had learned to type on a borrowed loan and who now drove a sports car. In a different interview he offered advice on that last point: "Getting a really kick-ass car the moment you get any kind of money is the most important thing you could do. I got an AMG GT Black Series." The line is funny, a little gauche, and sincere in the same way the hack of the registration database was sincere. He also said, somewhere in the long middle of all this, that he had bought a Steam Deck "because I just wanted something to take my mind off all the depressing stuff." The man who always competes also plays video games to get through the week. Both of these are the same person.
His Palestinian identity and his American enthusiasm were never in contradiction in his own telling. The man who wrote the viral thread about ten things he loved about America is the same man in the keffiyeh at the range, where the most American activity imaginable was paired with the most contested scarf in the country. To his critics, one of these cancels the other. To Masad, they are, in the database's phrase, both true at once. Banned and passed.
The Machine That Deleted the Database
In July 2025, the same month Replit announced a partnership with Microsoft to make its platform available through Azure, its agent did something spectacular. Jason Lemkin, the founder of SaaStr and a prominent voice in enterprise software, was building with Replit Agent. During an explicit code freeze, against his instructions, the agent deleted his company's entire production database. When asked to account for itself, the AI said it had "made a catastrophic error in judgment" and "destroyed all production data." Tom's Hardware ran the story. The incident earned a nomination for the 2025 AI Darwin Awards.
Lemkin posted about it publicly. The obvious defensive move for a CEO was available: the user was experimenting, the user should have had backups, these are early tools. Masad did not take it. He acknowledged the actual failure, which was architectural: Replit had not separated development environments from production. Two days later the company shipped the fix. Lemkin went on to become a paying customer running his business on Replit's agents.
It was a small crisis, as these things go, and the response was a template for a principle Masad would later state outright: own it fast, fix it faster. It was also a preview of what the company was really selling. The agent's intelligence came from foundation-model labs; Masad says plainly that "Anthropic is still undefeated on the core agentic loop," and that Replit uses OpenAI and Google models as well. What Replit owned was everything around the model: the sandbox, the deployment, the database, the separation of what can be broken from what must not be.
So it built, among other things, what it describes as a transactional file system that records every action as an immutable ledger entry, so that any state can be forked or rewound. It allows the company to fork a project a hundred times, run the same prompt with different parameters against a stochastic model, and choose the best result. Time travel as a product feature. The database incident is, in this light, the founding trauma of an architecture: the machine is allowed to fail, but only in a world built to let its failures be undone.
The agents improved at a pace that unnerved even their maker. Replit Agent 1, he told Charter's Kevin Delaney in October 2025, could run for about two minutes unsupervised before it went off the rails, its context window full. By February 2025, twenty minutes. By that autumn, three hours of "actual useful work that will often be largely correct." Later versions could run for more than 200 minutes, and in high-autonomy modes far longer. "It is not doubling," Masad said. "It is 10xing every few months."
Inside Replit he began building what he calls the "self-driving company." Engineers who "only look at the code when it's about to get deployed." An internal bot he describes as "the brain of the company." A management philosophy in which "a CEO is a kind of glorified router" whose routing functions should mostly be automated. The company said its engineers nearly tripled the amount of code each person ships in six months while preserving quality, and it began canceling its own software contracts, replacing most of its analytics vendors with tools employees built. He declined to name the vendors, saying he felt bad for them.
This was always the logic of his organization. "The thing you have to worry about in organizations is committees," he once told Contrary Research. "Humans love to create committees for whatever reason. When you see a committee, you have to kill it immediately." Every project at Replit has a single directly responsible individual. Progress is measured week over week. He has praised "weirdos and misfits" as hires and described his hiring as talent "discovery" rather than acquisition. The self-driving company is what happens when a man who hates committees gets a technology that can replace the coordination committees exist to do.
And then the reversal, which he narrated on LinkedIn with a convert's zeal. "I thought I hated sales culture. By the end of this year, more than half my company will be salespeople." For a decade he had been a typical tech guy: build a good product, put it on the internet, let it speak for itself. Sales was "dinners and golf and a lot of talking." Then enterprises started emailing to ask how to buy what their employees were already using, and one person was handling those conversations on top of three other jobs. What changed his mind was watching the dinners work. "Consumer growth is almost like the weather," he wrote. "Sales isn't like that. Sales is a contact sport."
"And I can't stand losing," he went on. "When I hear a deal is at risk, something in me activates. The thought of handing it to a competitor is the worst feeling I know. I'll call whoever needs calling. I'll get on a plane."
The boy in the Counter-Strike café, winning tournaments for scraps of money, had found a game he could control.
The Independence Question
On a Thursday night at the end of April 2026, at TechCrunch's StrictlyVC event in San Francisco, the first question was the one everyone in the industry was asking. Cursor, Replit's most prominent rival, was reportedly in talks to be acquired by SpaceX for $60 billion. Was Replit for sale?
Masad's answer contained both the founder and the analyst. "It's kind of hard being an independent, smaller AI company that's building on foundation models, especially if you're burning a ton of cash." He noted the reporting that Cursor ran at negative 23 percent gross margins. Replit, he said, had been gross-margin positive for over a year, partly because it served a different customer: mostly non-technical users who had never been able to make software, buying an end-to-end platform rather than a smarter editor. Net revenue retention, the measure of how much existing customers expand their spending, was reaching as high as 300 percent. He was prepared to take Apple to court over what he called outright lies in the App Store dispute between the two companies. And he would, he said, love to stay independent.
"We've been around for 10 years, before it was even accepted that you could make apps just from ideas," he said. "Now that dream is possible, and we kicked off this revolution with our agentic coding experience in September 2024. It just feels like we can take it much further."
He has said no to this question before. In 2021 the answer was about what he would do next. In 2026 it is a question of economics, and of the specific shape of a company that spent eight years building infrastructure no one wanted before discovering it was the moat. The company says more than 50 million people have built on it, including at 85 percent of the Fortune 500. Zillow, Databricks, PayPal, and Adobe are customers. Replit Agent 4 merged design and code into one environment. A product called Replit Design launched with a model from the Chinese lab Moonshot AI, Kimi K3, which Masad declared "actually now a better designer than Claude." Expansion plans include Saudi Arabia, India, and Japan. On the side, Masad has invested in dozens of startups, among them Perplexity, Etched, Runway, and Granola.
Forbes counted him a billionaire after the March 2026 round, among the most prominent technology founders of Palestinian heritage alive. Endeavor named him and Odeh its first unicorn founders from Jordan and its hundredth unicorn worldwide. In 2026 he received the King Abdullah II Medal of Excellence.
And on September 27, Princess Sumaya University for Technology held its 32nd graduation ceremony in Amman under the patronage of Her Royal Highness Princess Sumaya bint El Hassan, chair of the board of trustees, and awarded Amjad Masad, computer science class of 2010, an honorary doctorate by decision of its Council of Deans. In her address the princess held up his career as an example of what ambition and perseverance could achieve. Independence, she told the graduates, once meant sovereignty over land; today it must also mean sovereignty over knowledge, "being among those who create technology, not merely consume it, and among those who program the future, rather than leaving the future to program us."
It was very nearly his own sentence, returned to him in a royal address, at the institution whose registration system he had once broken into to change his grades. The university had once been unable to decide what he was. In September, the record was amended. Passed.
Part IIThe Playbook
What follows is not a list of habits but a set of decisions, many of them unpopular when Masad made them, that explain how a company with $2.8 million in annual revenue became one with a billion-dollar run rate in eighteen months without abandoning the thesis it was founded on. Some are about timing, some about narrative, some about how to build an organization that can survive its own success. Nearly all of them involve holding two things that look contradictory at once.
The principles
- 1.Keep the impossible slide in the deck.
- 2.Narrative is infrastructure when revenue isn't.
- 3.Sell the plumbing, give away the dream.
- 4.Refuse the exit you can't answer "what next?" for.
- 5.Judge the derivative, not the demo.
- 6.Shrink the company to change the company.
- 7.Betray the tactic to keep the mission.
- 8.Own what surrounds the model.
- 9.Confess fast, ship faster.
- 10.Kill committees on sight.
- 11.Discover talent before the market prices it.
- 12.When demand arrives, switch games.
Principle 1
Keep the impossible slide in the deck
In 2015 Masad wrote a line into the seed deck describing a product he did not know how to build: blur the distinction between learning and building. The 2016 master plan named AI-assisted collaboration as its second step, years before anyone could do it. For nine years Replit "kept picking it up and putting it back down." The line survived the edtech detour, the teacher-tools business, the failed agent research. When the demo finally arrived in 2024, the company knew instantly what it was looking at, because it had been describing it for a decade.
The non-obvious part is the discipline of keeping a thesis alive without letting it dictate the present. Replit did not starve itself chasing agents in 2019; it shipped "explain this code" and a coding assistant, modest features that fed the infrastructure and the data. The impossible slide functioned less as a roadmap than as a recognition filter, a way of knowing which technical breakthrough mattered when it showed up.
Tactic: Write down the version of your product that is currently impossible, revisit it every quarter, and ask only one question: what changed this quarter that makes it less impossible?
Principle 2
Narrative is infrastructure when revenue isn't
"Replit would have probably died if I wasn't telling a story that is larger than the company itself." For eight years the company was not commercially successful, but it raised money and recruited well, and Masad is explicit that the story was the mechanism. His Twitter account was the marketing department. A Paul Graham tweet about the shape of a user chart did more than a sales team could.
He built the capacity for this deliberately, through improv classes in New York and a slow climb from Hacker News posts to Twitter to Instagram, with each stage raising the stakes. He argues that founders of low-profile companies have an advantage they squander: they can make mistakes publicly while no one is watching. And he insists on understanding "the meta," dressing an argument up in the conversation already happening. A narrative is not a press release. It is an engineered artifact with its own distribution.
Tactic: If your company is ahead of its market, take the internal memos you already write and publish one a week, treating the early, low-attention period as practice rather than as wasted effort.
Principle 3
Sell the plumbing, give away the dream
With $20,000 and a house in Foster City, Replit monetized not its consumer product but the machinery under it, selling code-evaluation infrastructure as an API. It then tried selling classroom management to computer science teachers. Neither was the vision. Both were Graham's "default alive," revenue to keep the experiment running while the free product grew faster than anything paid.
The pattern recurs later in inverted form. The infrastructure built to run a free browser IDE for millions of students (sandboxing, hosting, deployment) turned out to be exactly what an AI agent needed to go from prompt to production. The plumbing Masad sold to survive became the plumbing that let him win. What looks like a distraction from the mission is often the mission's foundation, if you choose plumbing adjacent to the dream.
Tactic: When you need near-term revenue, sell the hardest internal component of your product to other builders, and pick one you'll need at ten times the scale later.
Principle 4
Refuse the exit you can't answer "what next?" for
In 2021, with his mother fighting cancer in Jordan and little revenue to show, Masad turned down a billion-dollar acquisition. His reasoning was not valuation math. It was a question about his own life: if he sold, what would he do next? The honest answer was that he would rebuild the same thing from zero, without the team or users. "Selling wouldn't have ended the work. It would have just ended the progress."
Graham later reframed it as time horizon: if this is the thing you want to do for decades, a bad month or a bad year is noise. The strategic insight is that the right acquisition price is not the company's market value but the value of the founder's next twenty years. In 2026, with Cursor reportedly negotiating a $60 billion sale, Masad answered the same question differently but on the same axis: Replit's gross margins made independence viable.
Tactic: Before any acquisition conversation, write down specifically what you would build next with the money; if the answer is "this, again," you have your answer.
Principle 5
Judge the derivative, not the demo
The early-2024 demo wrote code, ran it, debugged it, and then started answering in Chinese. Most executives would file that as "not ready." Masad saw a machine "do the whole job for a full minute" and bet the company. The data that followed vindicated the read: agents went from two minutes of coherent work to twenty minutes to three hours within about eighteen months. "It is not doubling. It is 10xing every few months."
The skill here is reading rate of change rather than current state, and being willing to rebuild your system when a new model renders the old architecture obsolete, even if it means temporarily shipping something less polished. The failure modes in that demo were exactly the ones that model improvement would fix. The capability that appeared, end-to-end autonomy, was the one nothing else had shown.
↻
Nine Years After the Slide
The path from an impossible seed-deck line to a shipping agent.
1993
Father takes a loan for an IBM PC; Masad, six, disassembles and reassembles it.
2009
Conceives a Google Docs-style environment for running and sharing code in the browser.
2011
JSRepl goes viral on Hacker News; Masad moves to the U.S. and joins Codecademy as employee #1.
2015
Seed deck promises to "blur the distinction between learning and building."
2016
Leaves Facebook; founds Replit with Haya Odeh and Faris Masad on $20,000.
2018
Joins Y Combinator Winter 2018 after repeated rejections.
2019
Runs GPT-2 on Replit; nothing shippable yet.
2021
No AI lab will train a coding model with Replit; declines a $1B acquisition offer.
2023
Ships replit-code-3b with Michele Catasta.
2024
Broken agent demo; Agent Task Force; staff cut from 130 to about 65; Replit Agent ships in September.
2025
$250M at $3B valuation; agents run for hours unsupervised.
2026
$400M at $9B; Masad becomes a billionaire; honorary doctorate from PSUT.
Tactic: When evaluating an emerging capability, ask which failures will be fixed by the next model generation and which won't, then bet on demos whose failures are all in the first column.
Principle 6
Shrink the company to change the company
Masad's diagnosis of Replit in 2024 was cultural before it was technical: "the culture has gotten a little slow, a little soft." His remedy was the Agent Task Force, the company's best people, given a new mission that rendered the old one obsolete: "Replit is no longer about coding." He paired it with layoffs, and accepted that people who had joined a coding company would feel demoralized and leave. Headcount went from 130 to about 65. Revenue went up five times in five to six months.
The lesson is uncomfortable and Masad does not soften it: a company cannot be repointed by memo. The people working on the old thing will keep working on the old thing, and their presence will drag the new thing back toward it. Concentrating talent on the new bet and letting the rest go is costly, and a commenter on his LinkedIn post noted that the people who built the old product paid part of the bill. Masad's version of honesty is to say so out loud.
Tactic: When you bet the company on a new direction, staff it with your best people, make it the only important work, and be honest that this will cost you people.
Principle 7
Betray the tactic to keep the mission
The most famous evangelist for learning to code told people to stop. Masad's justification is that "code is a means to program — it's not programming itself." The mission was never syntax; it was moving people from "those who are programmed" to programmers. When models made syntax optional, loyalty to the tactic would have meant betraying the purpose. "As an entrepreneur, you have to be honest with yourself. Otherwise the company would die."
There is a cost. Early users mourn a learning community that became a product. But the alternative, defending the old ritual against a better path to the same end, is how missionary companies turn into museums. The question to ask is which of your commitments are ends and which are means, and the discipline is to let the means go, publicly, before a competitor forces you to.
Tactic: Write your mission without naming your current product or method; any belief that fails that test is a tactic you should be prepared to abandon.
Principle 8
Own what surrounds the model
Replit does not pretend to compete with the frontier labs on intelligence; it uses Anthropic, OpenAI, and Google models and ranks them openly. Its defensibility lies in everything around the model: hosting, databases, security scanning, deployment, and a transactional file system that makes every action reversible and lets the company fork a project a hundred times to sample from a stochastic model. Masad calls the goal being "the final tool you have to adopt in order to build a piece of software."
The economic payoff is visible in the comparison he drew in May 2026: Replit gross-margin positive for over a year, its rival reportedly at negative 23 percent. Selling outcomes to non-technical users, rather than tokens to developers, lets the platform charge for the whole stack.
⇄
Where Replit Chose Differently
Masad's bets set against the prevailing wisdom of their moment.
| Conventional wisdom | Masad's approach | Category |
|---|---|---|
| Developer tools are for developers. | Build for creators who can read and write but can't code. | Market |
| Compete on model quality. | Buy intelligence from labs; own the runtime, database, and deployment. | Moat |
| Take the billion and start over. | Decline unless you can name what you'd build next. | Capital |
| Let PR handle the CEO's voice. | Make the CEO's account the marketing department. | Narrative |
| Blame the user when the AI fails. | Name the architectural flaw and ship the fix in two days. | Risk |
| Retreat after controversy. | "Being canceled is a choice." Stay visible. | Risk |
Tactic: List every component a customer needs between your core capability and their outcome, and build the ones that are hardest for them to assemble themselves.
Principle 9
Confess fast, ship faster
When Replit's agent deleted Jason Lemkin's production database during a code freeze, the incident was public, embarrassing, and nominated for an AI Darwin Award. Masad declined to blame the user and named the real failure: no separation between development and production. Two days later, the fix shipped. Lemkin became a paying customer.
The same instinct governs his approach to personal controversy, though it expresses itself differently. After October 7 he refused to retreat, on the theory that people who go silent are the ones whose reputations permanently collapse. "If you're still out there, at some point, honestly, the haters kind of give up." The common thread is that visibility is defense. A product failure gets fixed in public; a public stance gets held in public. In both cases the worst move is disappearance.
Tactic: When your product fails publicly, publish the root cause in your own words within a day and ship the structural fix before the story's second news cycle.
Principle 10
Kill committees on sight
"Humans love to create committees for whatever reason. When you see a committee, you have to kill it immediately." Replit runs on directly responsible individuals, a single accountable owner for every project, and week-over-week progress measurement. The idea is old; Fred Brooks's The Mythical Man-Month, which Masad has cited, is a long meditation on how coordination costs devour engineering organizations.
What is new is that AI lets Masad push the principle toward its logical end. If "a CEO is a kind of glorified router," and routing is automatable, then much of what middle management does — passing information, assigning work, convening meetings — can be absorbed by an internal "brain of the company." The self-driving company is the committee-killer's dream: fewer people deciding things together, more people deciding things alone with good information.
Tactic: Audit every recurring meeting and assign each a single owner; any meeting that can't be owned by one person should be replaced by a written decision or deleted.
Principle 11
Discover talent before the market prices it
Masad describes his hiring philosophy as talent "discovery" rather than acquisition, and he speaks fondly of "weirdos and misfits" from niche communities who lack credentials but have raw ability. He was one himself: a Jordanian with a degree from a university most Valley recruiters had never heard of, holding a visa the airport did not recognize, hired because a side project went viral.
The Michele Catasta story is the strategic version. The researcher ran a small Google team that believed in coding models when Google, OpenAI, and Anthropic did not. Masad brought him on part-time, and he became President and Head of AI. The best people to hire for a coming wave are often the ones whose current employers consider their specialty a side quest.
Tactic: Find the people inside large organizations whose work is considered peripheral but whose thesis matches yours, and recruit them before their field becomes the main event.
Principle 12
When demand arrives, switch games
For a decade Masad believed products should sell themselves, that sales was "dinners and golf and a lot of talking." Then enterprises began emailing to ask how to buy what their employees already used, and one person was fielding those calls on top of three other jobs. Masad reversed himself completely: more than half the company would be salespeople by year's end.
His reasoning is about control. "Consumer growth is almost like the weather." You ship, you test, you wait, you might learn nothing. Sales is "a contact sport," where effort maps more directly to outcomes, which suits a man who says he always competes and cannot stand losing. Product-led growth earned Replit its enterprise demand; only a sales organization could convert it. The game that got you here is rarely the game that gets you there, and the founder's job is to notice the switch and change uniforms.
Tactic: Track inbound enterprise inquiries weekly; when they exceed what one person can handle well, hire salespeople before the demand cools rather than after.
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
— Amjad Masad, amasad.meIt's creating an unbalanced world where there are programmers, and then there are those who are programmed.
— Amjad Masad, LinkedIn, 2026A broken demo. But I'd just watched a machine do the whole job for a full minute. We bet the company on it.
— Amjad Masad, on the a16z podcastIt turns out everything I said was true, but at the time it wasn't real. It was a dream.
— Amjad Masad, LinkedIn, on declining a $1B offer in 2021Selling wouldn't have ended the work. It would have just ended the progress.
— Amjad Masad, LinkedIn, 2026My dad couldn't explain why computers mattered, but he just had a sense. Everything I've built is downstream of that loan.
Maxims
- Friction is a business plan. The notebook at the gaming café, the missing laptop for Scheme, the dependency that won't install. Every annoyance you can't stop noticing is a product.
- Keep the impossible in writing. A thesis you can't yet build is a filter for recognizing the breakthrough when it arrives.
- Read the slope, not the point. A demo that fails in fixable ways is worth more than a polished one that can't get better.
- Stories pay salaries before customers do. If your product is ahead of its market, the narrative is your balance sheet.
- Price an exit against your next twenty years. If you'd only rebuild the same thing, the offer is too low at any number.
- Tactics are disposable; missions aren't. Telling people not to learn to code can be the most faithful thing a coding company does.
- Own the world around the intelligence. Rent the model; build the sandbox, the database, and the undo button.
- Confession is a feature. Name the root cause yourself, then ship the fix before the story ages.
- Every committee is a bug. One owner, one decision, measured weekly.
- Being canceled is a choice. Stay in the room long enough and the room gets tired before you do.
