Part IThe Story
Some years into the new century, a Yahoo registration page assembled four random letters for a new user. A program had stretched and warped them until no machine of the era could read them, and it presented them to her as a test of whether she was a person. The letters happened to spell WAIT. The woman on the other side of the screen did what the word said. Almost nothing else about her survives: not her name, not her browser, not the account she was trying to open. The only record is a joke told on a stage in Camden, Maine, in October 2009. She waited for twenty minutes, by the account of the man who designed the test, before she gave up. He liked to add that he was grateful the letters hadn't spelled RESTART.
It is a funny story, and it is also a slightly terrible one. A person was asked by a machine to prove she was not a machine, and she obeyed the instruction with a literalness no machine would have managed. Luis von Ahn told it at Pop!Tech in the self-deprecating register he uses for most of his inventions. Have you seen those distorted squiggly characters on web forms? Do they annoy you? I invented that. Strangers who learn what he made have, on more than one occasion, told him to his face: "Oh, my God, I hate you."
Thirteen years before the woman waited, an eighteen-year-old in Guatemala City had his own thing to prove: that he could read, write, and understand English well enough to be admitted to an American university. The test was the TOEFL, and that year there were no open seats in Guatemala. So the family paid more than $1,200 to send him to neighboring El Salvador to sit it. His mother was a physician who, by his account, spent nearly everything she had on his schooling. He passed and went to Duke. He did not forget the fee, the flight, or the arrangement in which a teenager's future depended on whether his family could afford an airplane ticket. Years later the word attached to that memory was extractive.
Hold the two examinations side by side: the one he was made to take, and the one he made the world take. Two hundred million times a day, at the CAPTCHA's peak, people somewhere typed squiggles into a box to prove their humanity, roughly ten seconds apiece. By 2025 the company he co-founded, Duolingo, reported $1.04 billion in revenue. It also sold an English proficiency test that costs $49, takes about 45 minutes, and can be taken at home. The arc between those points is not straight. It passes through a dissertation, a game about guessing words, a digitized archive of the New York Times, a green owl that the internet decided was a stalker, and a memo in April 2025 that briefly made him the most hated chief executive in consumer software. He insists he did not see it coming.
By the Numbers
The Arithmetic of Attention
$1.04BDuolingo revenue in 2025, up from $1 million in 2016
$414.1MDuolingo net income in 2025
130MMonthly active users (Q1 2025)
10.9MPaying subscribers as of June 2025
200MCAPTCHAs typed per day at their peak, ~10 seconds each
$49 vs. $250Duolingo English Test vs. a traditional proficiency exam
93%Share of Duolingo learners who, per von Ahn, never paid a cent
The Boy Who Wanted a Nintendo
Every account of Luis von Ahn's childhood begins with his mother, and so does his own.
She was one of the first women in Guatemala to complete medical school. She had her son at forty-two and raised him largely on her own, with help from her mother. "I was basically raised by two strong women," he told ABC News in 2022. The family on her side owned a candy factory, and the boy spent his early years taking in its machinery and trying to understand how the presses and conveyors worked. That instinct, wanting to see the gears behind the visible thing, would one day be called a research agenda. The family was not rich, though by Guatemalan standards it was not poor either. His mother made a deliberate, almost reckless decision about where its money would go. "She spent her entire wealth, really, her entire salary and everything on my education," he said. "So she basically gave me a rich person's education, even though we were not rich."
The rich person's education was the American School of Guatemala, a private English-language institution in the capital. Elsewhere he has estimated that a child in his country had about a 50 percent chance of learning to read above a first-grade level. By the World Bank's measure, nearly half the population lived in poverty. He has called the school "a great privilege," and he means it as an accounting term, a debt on the books.
When he was eight, his mother went out to buy him a present. He was sure it would be a Nintendo. It was a Commodore 64. "I was pretty upset," he remembered. "I guess it changed my life. I thank God she did that." The games for the machine cost money, and his mother would not hand it over. So the boy learned to program, built games of his own (he has called them "very crappy"), and started trading software with whoever would trade. Soon adults were ringing the doorbell of the house to make exchanges with a child. His mother shut the operation down. It was his first business, and his first lesson in the gap between something working and something being allowed to work.
The country around the house was at war with itself for most of his childhood. In 1996, his last year of high school, an aunt was kidnapped and held for a week. By his telling, that event pushed him to look beyond Guatemala for university. The American school had given him English. A math degree required leaving. Leaving required the TOEFL, and the TOEFL, that spring, required El Salvador.
Decades later, when his mother asked him whether he was all right for money, a question only a mother asks a billionaire, he answered patiently. "Yes, I am okay with money. Do not worry about it."
— Luis von Ahn, in an interview with Doug Rand, quoted by Carnegie Corporation of New YorkPeople talk about education as an equalizer, but I've seen how it can be a divider. When people are rich, they can pay for an amazing education, and because they get a great education, they remain rich.
How Lazy Cryptographers Do AI
He arrived at Duke in 1996 and graduated in 2000 with a mathematics degree, summa cum laude. That fall he started the computer science PhD at Carnegie Mellon in Pittsburgh, a city he had chosen almost entirely because of two people. In November 2025, standing on the CMU stage for the university's 125th anniversary, he thanked them by name. "I told a lot of people here," he said of Manuel and Lenore Blum, "they are the reason I am in Pittsburgh."
Manuel Blum was a University Professor of computer science, an honorific CMU does not hand out lightly, and a theorist of the strange zone where cryptography meets computational difficulty. Von Ahn did not intend to be an entrepreneur. "Ever since I was a teenager, I wanted to be a professor," he said at London Business School in 2022. His earliest serious work was as austere as computer science gets. With Nicholas Hopper and John Langford, he gave the first rigorous definitions of steganography in the computational setting, the art of hiding a message so an eavesdropper cannot tell a message is being sent at all, and proved that private-key steganography was possible. A concealment problem. Remember it.
Then, in his first year, the chief scientist of Yahoo came to campus with a list of ten problems the company could not solve. Von Ahn later admitted that nine of them eluded him. The tenth was this: automated programs were registering millions of free Yahoo email accounts, and spammers were using them to flood the world with junk. How do you tell, over a network, whether the thing filling in your form is a person?
The answer he and Blum reached has the cheekiness of the best theory. Some problems are hard for computers and easy for people, and reading distorted text was one of them. Turn the hardness into a gate. If a machine ever learns to pass the gate, then someone has advanced artificial intelligence, and the security failure is a scientific success. Heads, the spammers lose; tails, the field wins. The papers that followed carry titles that say a good deal about the young man writing them: "CAPTCHA: Using Hard AI Problems for Security," at Eurocrypt 2003, and, in Communications of the ACM in February 2004, "How Lazy Cryptographers Do AI." The acronym is a mouthful that has outlived its meaning: Completely Automated Public Turing test to tell Computers and Humans Apart. Yahoo deployed it, by the National Inventors Hall of Fame's account, in 2001. The rest of the web copied it almost at once.
"We made zero cents from this," von Ahn said. "We made no money. We were just very happy it was being used."
He collected CAPTCHA folklore the way other inventors collect patents, and he told it with evident glee. There was the woman who waited. There was a Slashdot poll asking readers to name the best computer science graduate school in America. CMU students wrote a program to stuff the ballot; MIT students wrote one back; within days both schools had more than six million votes and the poll came down. There were spammers who, unable to beat the test with software, hired people in low-wage countries to solve CAPTCHAs by hand. Von Ahn observed, with the dryness of a man who has done the math, that at least it was costing the spammers something, and at least it was creating jobs. There were pornographers who got their CAPTCHAs solved for free by showing the puzzles to visitors and paying them in pornography. There were Russian CAPTCHAs that asked users to compute the limit of an algebraic expression, and Indian ones that asked for the resistance of a circuit. American blog CAPTCHAs asked "What is 1+1?" He thought all of them failed, because all of them were easy for computers.
The pornographers in particular stuck with him, and that tells you something. The most cynical people on the internet had figured out what the most idealistic computer scientists had not: if you want people to perform a task a computer can't, you don't ask them. You trade them something they already want.
Stealing the Cycles
The arithmetic that defined his twenties went roughly like this. In 2003, people around the world spent something like nine billion hours playing solitaire. Computers could not look at a photograph and say what was in it. Most of the billions of images on the web had incomplete captions or none, which meant image search was poor and the visual web was closed to blind users. And humans, as the ESP Game's eventual documentation put it with a dryness that sounds like him, "are perfectly capable of it, but are not necessarily willing."
He and Blum talked about it in terms of theft, though they meant it kindly. If you could devise a game interesting enough, you could steal those brain "cycles," the term computer scientists use for units of processing, and put them to real work. The ESP Game went online in October 2003. The rules were austere. Two strangers are paired at random, unable to see or speak to each other. Both are shown the same image. Each types words describing it, and when the two type the same word, both score points and the word becomes a label. Two and a half minutes, fifteen images. The genius was in the constraints. Since the only thing the players shared was the picture, the only winning strategy was to describe the picture honestly. Vandalism was self-defeating. Once an image had been labeled enough times with the same word, that word became "taboo," forcing players to dig for the second-most-obvious description, then the third. Occasional test images with known answers caught anyone trying to game the system, and a label only counted once enough independent pairs had agreed on it. In the formal language of the analysis, the chance of corruption shrank exponentially with every repetition.
The 2004 paper reported that a pair of players produced about 3.89 labels a minute. At that rate, von Ahn and his co-author Laura Dabbish calculated, 5,000 people playing continuously could put a label on every image Google had indexed, 425 million of them, in 31 days. The site's launch page made the boast in its own voice: "If the ESP game is played as much as other popular online games, we estimate that all the images on the Web can be labeled in a matter of weeks!" It did not quite work out that way, but it worked out well enough. Some players logged more than forty hours a week. One journalist who tried it in 2007 confessed he was finding it hard to stop. Between launch and May 2008 the game collected 36 million labels. Google licensed it in 2006 and built its own version, Google Image Labeler, which ran until the Google Labs shutdown on September 16, 2011. Von Ahn has said the license brought him "a couple million dollars."
More games followed. In Peekaboom, one player slowly revealed parts of an image while the other guessed, which generated data on where in a picture an object sat. Verbosity harvested commonsense facts. Phetch produced descriptions meant to make images legible to the blind. His 2005 dissertation, simply titled Human Computation, was the first publication to use the term, which he had coined. It won CMU's School of Computer Science award for best doctoral dissertation. A tech talk he gave at Google on July 26, 2006, drew more than a million views, an absurd number for a lecture by a computer scientist about labeling.
That September, the MacArthur Foundation called. He was twenty-seven, the youngest of that year's fellows, and the grant was $500,000 with no strings. "Had I won the lottery, I probably would have bought a bigger house," he told CMU's news office. "But I think this money is meant for something more than that." Asked nearly two decades later by Wired what he did with it, he said: "I put it in the bank." It sat there until it helped pay to set up Duolingo. The other consequence of the genius grant was a running joke. "Every time you do something stupid — which happens all the time, with me — it's always the same joke: 'Haha, I thought you were a genius,'" he told NOVA. "But I do stupid stuff all the time."
The CMU press release from that fall preserves the scale of what the young professor wanted. The Panama Canal, he noted, took tens of thousands of workers. Apollo took 50,000. The pyramids perhaps 100,000. "Now, we can get 400 million people working together." What would they do? He did not know. "We just can't think big enough," he said. "I don't think I can think big enough yet."
His blog from those years shows the same mind with the brakes off. On March 5, 2009, he posted an idea filed under the label "Get Rich Quick Schemes." Computers can't reliably tell whether an image is pornographic. Image search engines need to block pornography. Many people like looking at pornography. "Everything aligns perfectly," he wrote. Let the people who like looking at porn label it, check their accuracy against images with known answers, and reward the accurate ones with better porn. When a commenter asked how "goodness" would be calibrated, he answered within the hour: eleven choices, NOT PORN or PORN 1 through 10. He was never quite brave enough to launch it. The post is a joke and also the purest statement of his method anywhere in the record. Find the want. Attach the work to it. Check the answers.
The Guilt of the Inventor
There is a version of the reCAPTCHA story told as a eureka, and von Ahn has told it that way. The version he returns to more often is about guilt.
By 2007, by MIT Technology Review's estimate, people around the world were solving some 60 million CAPTCHAs a day. Later his own figure was 200 million. Each took about ten seconds. "At first I was feeling quite proud of myself," he told the magazine that year. "But then I was feeling bad, because every time you solve a captcha, you waste 10 seconds." At Duke in 2009 he gave the same confession a punchline: "That meant a huge chunk of human hours were wasted typing my CAPTCHAs. Then, I started feeling bad." The journalist Alex Hutchinson, writing in The Walrus that March, put it more grandly: von Ahn had realized he "had unwittingly created a system that was frittering away, in ten-second increments, millions of hours of a most precious resource: human brain cycles."
The fix borrowed the ESP Game's logic and pointed it at a problem with an obvious owner. Book digitization projects were scanning pages and running them through optical character recognition software, which failed on a meaningful share of words. By one estimate he cited, OCR couldn't read about 30 percent of the words in books printed before 1900. Those failures were exactly the kind of distorted text only people could read. So reCAPTCHA, released on May 27, 2007, showed every user two words. One was already known and served as the control. The other was a word the scanners had choked on. Type the control correctly and the system assumed you had probably read the unknown word correctly too. Each OCR program's guess counted for half a point and each human answer a full point, and at 2.5 points a reading became canonical and could serve as a control itself. If six people rejected a word as unreadable before anyone agreed on it, the system threw it out. The arrangement is the ESP Game with the fun removed and replaced by necessity. You had to solve the CAPTCHA anyway. Now the ten seconds went somewhere.
Brewster Kahle, the co-founder of the Internet Archive, an early partner, offered the line that ended up in every profile: "It is like an army of ants building the Taj Mahal." The New York Times signed on to digitize its archives. Craigslist adopted it under the slogan "Fight Spam, Digitize Books." So did Facebook, Ticketmaster, Twitter, CNN.com, and the federal website for coupons toward digital TV converter boxes. The company never needed venture capital, though by von Ahn's account plenty of venture capitalists offered. Google acquired it in September 2009 for an undisclosed price he has since described as "in the tens of millions of dollars." It never left Pittsburgh. By 2010 it was serving more than 100 million CAPTCHAs a day. By the Hall of Fame's count, more than a billion people eventually helped digitize text through it. "We were probably doing a year of the New York Times in a week," he told Forbes Africa.
Every story about reCAPTCHA arrives with its shadow, and an honest account has to include it. The moral calculation that made reCAPTCHA feel virtuous, that the labor was happening anyway, also made it unpaid labor at enormous scale, done by people who were mostly not told. A Massachusetts woman sued Google in 2015 on exactly that theory. After the acquisition, the puzzles shifted from book scans to Street View photographs of house numbers, crosswalks, and traffic lights, and the question of who benefited grew murkier. A 2023 academic study estimated that reCAPTCHA had cost humanity some 819 million hours of unpaid time. Google shut down the original, book-digitizing version on March 31, 2018. Today reCAPTCHA is mostly invisible, a background score assigned to your behavior, and it now sits under Google Cloud's fraud-defense product line.
When Stephen Dubner asked him on Freakonomics in 2020 what Luis von Ahn and Tom Sawyer had in common, von Ahn didn't hesitate. "I'm going to guess. It probably has to do with tricking somebody to paint a fence." Dubner, polite, offered "entice." The distinction is the whole argument, and von Ahn, to his credit, let the word "trick" stand.
Translating the Web, and Other Abandoned Purposes
In 2009, freshly enriched by his second sale to Google before thirty, a tenured-track professor with a MacArthur in the bank and nothing he had to do, von Ahn asked himself a question he later wrote down in a letter to Duolingo's employees: "What kind of world do I want to live in and how can I make the greatest impact in creating that world?"
The answer involved a graduate student. Severin Hacker was Swiss-born, an immigrant to Pittsburgh by way of the same PhD program. By von Ahn's account he was a standout student, and he believed, in his own words, that "free education will really change the world." He later had a hand in one of the most consequential branding choices in the history of educational software: the company mascot is a green owl because Hacker hates the color green. Asked during a 2012 Reddit AMA which fictional duo he and Hacker most resembled, von Ahn answered: Beavis and Butthead. Hacker became chief technology officer and has stayed there. The National Inventors Hall of Fame later recognized him too.
Their idea, in its first form, was pure von Ahn. It was a two-sided human computation machine. People wanted to learn languages, and 1.2 billion of them were trying, two-thirds learning English, most of them, he believed, in order to earn more. The web wanted translating. When he looked at the Spanish internet his relatives in Guatemala used, he saw a poorer place. "It's got much less information," he told the Times in June 2012. "I see people struggling with that a lot. They don't get the information we take for granted." So beginners would get lessons, and as they advanced they would translate real sentences from real web pages, voting on one another's attempts. The translations would be sold, or given away, to the sites that needed them. "You're learning a language and at the same time, helping to translate the Web," he said. "You're learning by doing."
It was CAPTCHA turned inside out. reCAPTCHA had attached useful work to a chore. Duolingo would attach it to a desire.
The project ran first on the MacArthur money and a National Science Foundation grant, and the founders seriously considered making it a nonprofit. Von Ahn decided against it. Charities have to ask for money forever, and he judged the model unsustainable. In April 2011 he gave a TEDxCMU talk, "Massive-scale online collaboration," that has since been played nearly two million times. He has credited that audience with giving Duolingo its first base of users. In October 2011 the company raised $3.3 million in a Series A led by Union Square Ventures, with Tim Ferriss and Ashton Kutcher's A-Grade Investments joining. The private beta opened at the end of November. By December 13 the waiting list topped 100,000. When Duolingo opened to the public on June 19, 2012, the list was around 500,000. Before launch, von Ahn later wrote, "I remember thinking it would be pretty cool if one million people ever used our website."
A $15 million Series B led by New Enterprise Associates followed in September 2012. An iPhone app arrived in November and Android in May 2013, when the company had about three million users. That December, Apple named Duolingo its iPhone App of the Year. In February 2014, Kleiner Perkins led a $20 million Series C. The company then had 34 employees, around 25 million registered users, and essentially no revenue.
The translation business never became what it was meant to be. It's tempting to call this a failure, and in one sense it was. The animating mechanism, the thing that made Duolingo feel like his kind of idea, was gradually retired, the way the old reCAPTCHA was. By June 2015, when Google Capital led a $45 million round at a valuation of about $470 million, TechCrunch reported that the company was leaning on its test center certification as a revenue source and had "other plans to monetize." The crowdsourced language course incubator, launched in October 2013 and staffed by volunteers who built courses in Irish, Navajo, Hawaiian, Esperanto, and Klingon, was shut down in March 2021 in favor of professional linguists. The community forums closed in March 2022. One by one, the parts of Duolingo that depended on the crowd as labor were swapped out for parts that treated the crowd as customers.
Asked once in an AMA what everyone should know about Duolingo, he answered in four words: "That it works!" Asked about the owl, he said it beat the alternative: "a broken stone with foreign characters on it."
The Toll Booth in El Salvador
The business that eventually rescued Duolingo's model arrived, as von Ahn tells it, by email. "We got a lot of emails saying thank you," he told The Guardian in August 2014, "but in order to get a job I need to certify" that I speak English. The learners had learned. Now they needed proof.
He already knew the proof business from the inside. It was the business that had put him on a flight to San Salvador at eighteen. Traditional English certification exams cost around $250, a month's salary in much of the developing world, by the Lemelson-MIT Program's accounting, and they took about three hours at testing centers that might be in another city or, in his case, another country. The Duolingo Test Center, launched in 2014, was a phone app and cost $20. When he announced it in a November 2014 Reddit AMA, he did not use the language of an investor deck: it was "created to replace expensive standardized English tests that take advantage of students and job-seekers with absurd prices, like the TOEFL and IELTS."
It grew into the Duolingo English Test: computer-adaptive, scored from 10 to 160, about 45 minutes, taken at home, priced at $49. Questions adjust to the test-taker's ability in real time, using machine learning and natural language processing the Duolingo team developed. The IEEE, in awarding him its 2026 Frances E. Allen Medal, called it "the first high-stakes proficiency exam to be delivered entirely online." By the IEEE's count, more than 4,200 institutions accept it.
Look at the shape of the move. Every business model Duolingo has run has charged the people with money so the people without it could keep learning for free. The translation model would have charged websites. The test charges a fee, a small one by the standards of its rivals, at the exact moment a learner's investment pays off: a visa, an admission, a job. Subscriptions charge the minority who want no ads, unlimited retries, and, more recently, an AI conversation partner. When the company passed a million paying subscribers in April 2020, its content was still free. "At least on Duolingo," von Ahn wrote that month, "more money can't buy you a better education."
That claim is about a curriculum, not a life, and he would presumably concede the difference. But the test changes the life. It is the one place in the Duolingo system where the company acts as a gatekeeper and not a teacher, and he priced the gate as though he remembered standing in front of one.
Making Learning as Addictive as Everything Else
Here is the paradox at the center of the most successful education app ever built. It was designed by a man who believed people would not, on their own, do what was good for them.
He had spent a decade studying exactly that. Humans are capable of labeling images, he had written, but not necessarily willing. Language learners are the same. In a 2013 Reddit AMA, asked why other language software failed, he wrote: "Because they don't motivate you to keep going! We spend a lot of effort making Duolingo addictive — we think the hardest part about learning anything on your own is motivation." Asked in 2017 why Duolingo didn't let learners chat with other learners, he gave an answer that is part data and part confession. Dozens of systems offered it, he said, and the most popular had less than 0.1 percent of Duolingo's active traffic. "I believe it's because most people just don't have the guts to talk to strangers in a language that they're not very familiar with (I know I don't!)." He has also said that the people who learn languages best "have no trouble sounding stupid." The two statements fit together. Most people mind sounding stupid very much, and he built a product for them.
What he built was a game with a purpose, now aimed at the player himself. Short lessons. Experience points. Leagues of up to thirty strangers ranked weekly, through Bronze, Silver, Gold, Sapphire, Ruby, Emerald, Amethyst, Pearl, Obsidian, Diamond. And above all the streak, a count of consecutive days practiced, symbolized by fire and guarded by notifications. Those notifications are chosen by what Duolingo's researchers described in a 2020 paper as a "sleeping, recovering bandit algorithm," a learning system that tests reminder after reminder to find the one most likely to bring each user back. Some of the sentences learners translate are deliberately absurd: The bride is a woman and the groom is a hedgehog. That reportedly follows from psychological research suggesting that surprising sentences are easier to remember. It is the ESP Game's taboo words again, refusing the obvious answer so the brain has to work.
He has always treated pedagogy as an empirical question, which is a professor's habit with a billion-dollar data set behind it. "If we want to know whether we should teach plurals before adjectives," he explained in 2014, "for the next 50,000 users, we'll teach half of them plurals before adjectives, half the other way around and then we measure which of these groups learns better. We figure out in a couple of days which is more effective and then switch everyone to that." The data produced findings he reports with visible delight and no explanation: "For some reason, Italian women learn English much better than Italian men." In conversation with Eric Ries he has said Duolingo spent its first five years focused on improving retention, which is to say, on making people come back.
The independent evidence on whether it teaches is mixed in the way such evidence usually is. A 2017 study found elementary students learning Spanish through the app did about as well as peers in classrooms. A 2022 paper in Foreign Language Annals, co-authored by Duolingo researchers, found that adults who finished its beginner courses reached reading and listening proficiency comparable to university students after four semesters. A 2023 study funded by the company found English learners didn't pick up much grammar. Critics, including some professional linguists, say the game design invites strategies that conflict with learning: chasing points, protecting streaks, gaming the leagues. Kotaku once asked why a language app was "plagued with cheaters." Von Ahn, who designed the ESP Game's anti-cheating test images, would not be surprised.
The owl did the rest. Duolingo grew, von Ahn told the BBC, "solely due to positive word of mouth" until 2019, with no advertising at all. The internet did his marketing by turning Duo into a menace, a green bird depicted in countless memes stalking users who skipped their lessons. The company leaned in. Its April Fools' video in 2019 introduced "Duolingo Push," a man in an owl costume who follows users until they practice. In 2024's "Duo on Ice," the owl admits a taste for human flesh. On February 11, 2025, the company announced that Duo had died after being hit by a Tesla Cybertruck. Users then had to pool 50 billion experience points to resurrect him. Duo's brand guidelines specify that he is a boy and communicates only in text. His full name, the company has said, is Duo Keyshauna Renee Lingo, and he was born in 1000 BC. Next to the Pittsburgh headquarters, Duolingo runs a taco stand, Duo's Taqueria, where customers are encouraged to order in Spanish. It brought in $700,000 in 2023.
Underneath the jokes is the old trade, now aimed at people instead of pixels. In April 2023, von Ahn gave a second TED talk, played nearly 13 million times. It was titled "How to make learning as addictive as social media." On a Stanford stage in 2025 he gave the plainest version of the goal: "I want it to be the case that we can show that screen time is actually useful for the world." The ambition is admirable. It is also an admission that he thinks the screen, not the classroom, is the competition.
A Professor Learns to Run a Company
He was not, by his own account, good at it at first.
At London Business School in November 2022, he described his management of Duolingo's first thirty employees as "micromanaging the hell out of" them. Asked by the BBC in 2021 what he wished he had learned earlier as a CEO, he named something no dissertation teaches: "one thing I wish I'd concentrated on earlier was my social skills." On Tim Ferriss's podcast in 2016 he said he wished he'd had more formal management training at Carnegie Mellon. Business Insider later summarized his evolution in a phrase he seems to enjoy: from micromanager to mascot. In 2020, when Duolingo had about 250 employees, he wrote that its retention rate exceeded 95 percent.
He also stayed in Pittsburgh, and that was a choice he defended in public long before it looked smart. On June 7, 2010, after selling reCAPTCHA without leaving town, he wrote a blog post explaining why. CMU's students were world-class. The competition to hire them was a fraction of what it was in California, "since it's not the case that everybody and their mother has a startup in Pittsburgh." The last time he visited Silicon Valley, he wrote, the guy who served him at Starbucks overheard him talking to a venture capitalist and started pitching. "It always strikes me when I go there how much everybody thinks exactly the same as each other (most even dress the same as each other)." His conclusion: "If I were to start another company, I would do it out of CMU." The following year he did. Duolingo became, by CMU's reckoning, the first Pittsburgh-based tech startup valued at more than $1 billion. That happened in December 2019, when Alphabet's CapitalG put in $30 million at $1.5 billion.
At the end of 2020, a new kind of investor arrived. Henry Ellenbogen is, in his own way, another child of a single mother. He was born in 1973 to a Spanish father and an American mother, who took her three children from Madrid to Miami when he was six and raised them on a nursing home administrator's salary. He is also the grandson of a Viennese immigrant who came to Pittsburgh in 1919. That man worked the floor at Kaufmann's department store by day and attended Duquesne Law School by night. He passed the Pennsylvania bar with the highest marks ever recorded, became a New Deal congressman, and quietly answered 44 boxes of letters from European Jews begging for help escaping before the war. He is buried in West View Cemetery in Pittsburgh. His grandson built one of the great records in growth investing, first at T. Rowe Price, then at his own firm, Durable Capital Partners, on the conviction that the people running a company matter as much as the model.
He met von Ahn on a Zoom call in 2020. When it ended, a colleague started listing the business's virtues, and Ellenbogen cut him off. "He reminds me of Tobi from Shopify," he said. "We're clearing our schedule. Whenever Luis will spend time with us, we're going." They met next in Duolingo's Pittsburgh office. It was the middle of the pandemic, and the building was empty. Von Ahn had been told that a private company thinking about going public should talk to Henry Ellenbogen. Beyond that, he didn't know what to expect. Ellenbogen talked about patterns. "He just seemed like a savant at identifying how companies work," von Ahn later told Colossus. Von Ahn, for his part, made an admission that should be required reading for every founder with a doctorate: "I'm going to be a first-time CEO. My sense is you're going to see things based on your experience that I haven't seen yet."
Durable led Duolingo's last private round in November 2020, at $2.4 billion. The company filed to go public on June 28, 2021, under the ticker DUOL, and when it listed that July the shares closed up 36 percent on the first day, putting the company's value near $5 billion. Ellenbogen bought more at the IPO. Within months the stock started to slide, from around $200 to a low near $70 in 2022, as rising interest rates punished every company valued on distant profits. That year Ellenbogen took von Ahn and his chief financial officer, Matt Skaruppa, to dinner in New York and told them what he believed was the real story of Amazon. In his telling, Amazon did not ignore profits to grow. It nearly died, then ran its core business with obsessive discipline, and only then spent on new bets. The era of free money was over, he told them. Prove you can make money.
"It was very good advice," von Ahn said later. "It probably made our company worth three times more because we increased profitability and the market rewarded it." Revenue, $250.77 million in 2021, reached $531 million in 2023. In 2025 it crossed a billion, with $414.1 million in net income. In July 2025, as Figma filed to go public, von Ahn joined its board. Ellenbogen had made the introduction.
The first-time CEO had become someone you call before an IPO. He has also made unpopular product decisions and kept them. In August 2022 Duolingo replaced its branching "tree" of lessons, which let learners choose where to go next, with a single linear path. Users revolted on Reddit and Twitter, and von Ahn said there were no plans to reverse it. In October 2023 the company stopped updating its Welsh course to "focus on languages in higher demand," over objections tied to the Welsh government's language revival strategy. In 2026 it announced that Duolingo for Schools would be shut down in 2027. Each decision traded something for focus. Each one cost goodwill. The goodwill, until April 2025, kept coming back.
AI-First
On April 28, 2025, von Ahn sent an email to all Duolingo employees and posted it on LinkedIn. It was meant, presumably, as the kind of internal memo that gets shared because it is bold. Duolingo would become an "AI-first" company. It would gradually stop using contractors for work AI could handle. Teams could add headcount only if they couldn't automate more of their own work. The line that traveled farthest was a preference: the company would rather "take occasional small hits on quality than move slowly and miss the moment." Within days Duolingo announced 148 new language courses built with generative AI, more than doubling its catalog.
The response was quick and furious, and it came from the people who had made the owl famous. "This is a disaster. I will cancel my subscription," read one comment Fortune collected. "AI first means people last," said another. Users posted screenshots of broken streaks, years of consecutive practice deliberately ended in protest. "That owl is dead to me," one headline quoted. In mid-May the company deleted every post from its TikTok and Instagram accounts, the most celebrated social media presence in corporate America, which had grown out of an owl that threatened to eat your head. It came back with a video of someone in a three-eyed owl mask ranting that "everything came crashing down with one single post about AI."
The backlash was not only about contractors. Around the same time, a remark from the May 8 episode of the No Priors podcast was circulating. "Ultimately, I'm not sure that there's anything computers can't really teach you," von Ahn had said. "It's just a lot more scalable to teach with AI than with teachers." Teachers wouldn't disappear, he added; schools would still be needed, because "you still need people to take care of the students." Many listeners heard a man reducing the classroom to childcare. Months earlier, Wired's Katie Drummond had asked him what job AI should never do. "A lot of jobs," he had said. "I think that anything where humans need to be inspired, like teachers. Humans need to be inspired. It's kind of hard to get inspired by AI."
Both statements are on the record. They do not cancel each other, exactly. One is about instruction and the other about inspiration, and he has always believed motivation is the scarce resource and information the plentiful one. But the gap between them is where the anger went.
Within a week he posted a clarification: he did not "see AI replacing what our employees do," but rather a "tool to accelerate what we do, at the same or better level of quality." In June he told the Financial Times: "I did not expect the amount of blowback." He added a defense that was both true and impolitic: "Every tech company is doing similar things, [but] we were open about it." By August he was more contrite with the New York Times: "This was on me. I did not give enough context. We've never laid off any full-time employees. We don't plan to." Contractors had always gone "up and down depending on needs," he said. He introduced "F-r-A-I-days," weekly time for teams to experiment with AI. In January 2024, the company had cut about 10 percent of its contractors with the explanation that "generative AI is accelerating our work by helping us create new content dramatically faster." That earlier round had passed with less notice.
On the August 2025 earnings call, analysts asked about the backlash. He answered as the professor who always checks the numbers: "We just don't believe that the effect of this is very material in terms of when you're looking at actual financial metrics." By the financial metrics, he was right. Revenue crossed a billion dollars that year. Paying subscribers had reached 10.9 million by June. The chess course, launched that April and reportedly "vibe coded" into a hit, had seven million daily active users by early 2026.
The episode reads differently against the arc of his career. He built his name on a single question: what can humans do that computers can't? CAPTCHA was a gate at that boundary. The ESP Game and reCAPTCHA were mines dug along it. Duolingo began as an engine running on the assumption that only people could translate the web. "Language does not come naturally to machines," the Times wrote in 2012, introducing a man who was "betting on" the human edge in irony, wordplay, and cultural context. In October 2023 researchers found that GPT-4 could solve CAPTCHAs. The boundary had moved. He moved with it, faster than the people who had played his games for twenty years wanted him to. Some of them had only ever experienced him as a set of squiggly letters asking whether they were human.
— Luis von Ahn, to the Financial Times, June 2025I did not expect the amount of blowback.
What the Machine Still Can't Do
By now his story gets told as uplift: the immigrant genius, the "genius grant," the inventions everyone uses, the slim biography for young readers, Luis von Ahn: Innovating Education and Technology for a More Accessible World. The honors pile up. The Grace Murray Hopper Award in 2011. The Presidential Early Career Award in 2012. A $500,000 Lemelson-MIT Prize in 2018. Carnegie Corporation's Great Immigrants Award in 2021. Induction into the National Inventors Hall of Fame in 2023, under U.S. Patent No. 8,555,353. The IEEE's Frances E. Allen Medal for 2026. In 2021 he joined the executive committee of the Partnership for Central America, convened in response to Vice President Kamala Harris's call to invest in the region migrants were leaving. That same year he and his wife Ingrid, a Swedish national, started a foundation to support Guatemalans, especially women and girls, giving out $3 million in 2022. He is learning Swedish on the app. He told Wired he needs to get better at it.
The uplift version is not wrong. It just leaves out the parts that make him interesting. He is a man whose great insight was that human attention is a resource you can harvest, and he built a business, eventually, that harvests it with streaks and leaderboards and an owl. He started with a nonprofit's instincts and ended with a public company's margins. His career has been organized around the question of what people can do that machines can't, and he says AI can teach anything. He also says it can't inspire anyone. He has said he was once motivated by "surprising, non-obvious, clever ideas" and is now motivated "by large-scale impact." Both motives are in everything he has made, and so are both reactions: I hate you, and I love you. Fortune reported in 2019 that the second phrase was what people said when they learned he had built Duolingo.
"We've been able to educate literally millions of people," he told the Hall of Fame. "Most of them, in fact, 93% of them, have not paid us a single cent. And yet we were able to make a very successful company out of that. I'm very proud of that."
Near the bottom of the Lemelson-MIT Program's summary of his work is a list of the institutions that accept the Duolingo English Test for admission. It is the test that costs $49, lasts about forty-five minutes, and can be taken on a computer at home, in Guatemala City or anywhere else, with no plane ticket. The list includes UCLA, Columbia, Dartmouth, NYU, Yale. The first name on it is Duke.
Part IIThe Playbook
What follows is less a biography than a set of mechanisms. Von Ahn's career is unusually legible because he has explained most of his moves in public, often within hours of making them, sometimes on Reddit. The principles below come from that record: from CAPTCHA, the ESP Game, reCAPTCHA, the slow reinvention of Duolingo's business model, and the AI controversy that tested all of it. Several are uncomfortable. The best ones usually are.
The principles
- 1.Find the wasted cycle.
- 2.Make the selfish move the useful one.
- 3.Give away the first breakthrough.
- 4.Hold the mission fixed and let the business model be a hypothesis.
- 5.Charge at the moment of payoff, where incumbents extract.
- 6.Compete with the phone, not the classroom.
- 7.Let the experiment settle the argument.
- 8.Build outside the monoculture.
- 9.Stay deliberately under-monetized.
- 10.Name the skill you lack, then rent it from someone who has it.
- 11.Hold the product line; never skip the context.
- 12.Let the brand misbehave.
⏱
From Ten Seconds to a Billion Dollars
The chain of inventions, each built on the logic of the one before.
1996
Flies to El Salvador to sit the TOEFL; enters Duke at 18.
2000
After a Yahoo talk on bots, co-creates CAPTCHA with Manuel Blum. "We made zero cents."
2003
The ESP Game goes online; 36 million image labels by May 2008.
2005
PhD thesis, Human Computation, coins the term.
2006
Google licenses ESP as Image Labeler; MacArthur Fellowship at 27.
2007
reCAPTCHA launches, turning CAPTCHAs into book digitization.
2009
Google acquires reCAPTCHA; Duolingo conceived with Severin Hacker.
2012
Duolingo opens to the public with ~500,000 on the waitlist.
2014
Test Center launches at $20, the seed of the Duolingo English Test.
2021
IPO on Nasdaq; volunteer course program ends.
2025
"AI-first" memo and backlash; revenue crosses $1 billion.
Principle 1
Find the wasted cycle
Every major von Ahn project starts with the same accounting: human effort that is already being spent, and that produces nothing. Nine billion hours of solitaire. Two hundred million CAPTCHAs a day at ten seconds apiece, which he multiplied out to half a million hours of human attention daily. The ESP Game did not ask anyone to label images. It noticed people were already playing games and gave the play something to produce. reCAPTCHA did not ask anyone to transcribe books. It noticed people were already being forced to read distorted text.
The discipline is to look for effort before looking for demand. Most founders start by asking what people want. Von Ahn starts by asking what people are already doing that could be made to count. The advantage is that you skip the hardest part of any product, behavior change, because the behavior already exists. You are adding a byproduct, not a habit.
Tactic: Before designing a new behavior, list the repetitive tasks your users already perform and ask what each one could produce as a side effect.
Principle 2
Make the selfish move the useful one
The ESP Game works because the only way to win, matching a stranger's word when all you share is an image, is to describe the image accurately. Lying is a losing strategy. Taboo words keep players from coasting on the obvious answer. Test images with known labels catch cheaters, and requiring agreement across independent pairs drives the error rate down exponentially. The "Porn or Not" proposal on his blog, never launched, is the same structure with the politeness removed: reward accuracy with the thing the participant actually came for.
This is mechanism design dressed as entertainment, and the lesson for operators is that incentive structure beats exhortation. Von Ahn learned it partly from the spammers and pornographers who were solving CAPTCHAs at scale by paying people in what they wanted. The cynics got there first. He just aimed the method at better ends.
Tactic: Design every system so that the easiest way for a participant to get what they want is also the way that produces the output you need, and verify with a few known answers.
Principle 3
Give away the first breakthrough
CAPTCHA earned its creators nothing. "We made zero cents from this," von Ahn has said, without regret: "Everyone started using it; all kinds of people started approaching me." Free distribution made CAPTCHA universal, and universality created both the problem (the wasted half-million hours) and the platform (millions of daily solves) that reCAPTCHA monetized. Duolingo repeated the pattern on a larger scale. In 2020 he named "give the product away for free" as one of the company's two best decisions, because it "allowed us to scale rapidly, which has given us access to unprecedented amounts of data on how people learn."
The pattern works when the free thing produces an asset that compounds: a standard, a data set, a reputation. It fails when the free thing is merely a discount. CAPTCHA and the free Duolingo curriculum both produced data and distribution that later products could capture.
Tactic: Give away the version of your breakthrough that generates data or distribution, and save monetization for the second product that the first one makes possible.
Principle 4
Hold the mission fixed and let the business model be a hypothesis
Duolingo's founding business model, selling translations produced by learners, was elegant, original, and essentially abandoned. What survived was the mission: free, high-quality language education. The revenue engine went through a series of experiments: translation, a $20 test, advertising, subscriptions, a premium AI tier. The founders considered a nonprofit and rejected it as unsustainable. In conversations since, von Ahn has credited an investor with pushing him and Hacker to monetize at a time when the company was worth a great deal on paper and earning almost nothing.
The discipline is emotional as much as strategic. The translation model was the most "von Ahn" thing about Duolingo, the direct heir of reCAPTCHA. Letting it go meant giving up the cleverness that had defined his career for something more ordinary that worked.
$
Duolingo's Revenue Experiments
The mission stayed constant; the business model was rewritten repeatedly.
| Model | Who pays | Outcome |
|---|---|---|
| Crowdsourced web translation | Websites needing translation | Retired |
| Test Center / Duolingo English Test | Learners at the moment of certification | Scaled |
| Advertising | Advertisers, against free users | Supporting |
| Super Duolingo subscription | The paying minority | Core engine |
| Duolingo Max (generative AI) | Premium subscribers | Expanding |
Tactic: Write your mission as a constraint the business model must satisfy, not a description of the business model, so you can replace the model without betraying the mission.
Principle 5
Charge at the moment of payoff, where incumbents extract
The Duolingo English Test is the most personal product von Ahn has built, and probably the most strategically precise. It charges learners at the exact moment their learning becomes valuable: an admission, a visa, a job. It enters a market where incumbents charged around $250 for a roughly three-hour exam at a physical center. Von Ahn himself once paid more than $1,200 to reach such a center. He priced Duolingo's version at $49, taken at home in about 45 minutes. More than 4,200 institutions now accept it.
The lesson is twofold. First, the best time to charge is when the customer can see the return. Second, the best market to enter is one where the incumbent's price reflects its control of a gate rather than its costs. Von Ahn called the testing industry extractive. Extractive incumbents are vulnerable to anyone willing to make less money per customer from far more customers.
Tactic: Find where your customers already pay a toll they resent for a credential they need, and offer the credential at a price that makes the old toll look indefensible.
Principle 6
Compete with the phone, not the classroom
Von Ahn decided early that the hardest part of learning on your own is motivation, not information. "We spend a lot of effort making Duolingo addictive," he wrote in 2013. His 2023 TED talk was titled, without irony, "How to make learning as addictive as social media." Streaks, leagues, notification algorithms, a menacing owl: these are the tools of the attention economy turned toward self-improvement. His rival was never the language school. It was every other app on the home screen.
This framing explains decisions that look strange from the outside. The linear lesson path replaced a branching tree despite user anger. Learner-to-learner chat was rejected because almost nobody used it. Absurd sentences about hedgehog grooms are used because surprise aids memory. The cost is real: critics argue that gamification rewards point-chasing over learning, and cheating is common enough that gaming press has covered it. Von Ahn accepts the trade, because an app nobody opens teaches nothing.
Tactic: Identify the most compelling product your users touch every day, whatever its category, and benchmark your engagement mechanics against it rather than against your nominal competitors.
Principle 7
Let the experiment settle the argument
"If we want to know whether we should teach plurals before adjectives," von Ahn explained in 2014, "for the next 50,000 users, we'll teach half of them plurals before adjectives, half the other way around and then we measure which of these groups learns better." This is a professor's reflex given a scale no university has. The company's own published work treats notification timing, sequencing, and curriculum design as testable. The findings are sometimes inexplicable. "For some reason, Italian women learn English much better than Italian men." He reports them anyway.
The point is not that data replaces judgment. Von Ahn kept the linear path against loud objections. The point is that a large user base lets you replace opinion with measurement on most questions, which saves judgment for the few questions experiments can't answer.
Tactic: For every recurring internal debate about sequencing or design, ask whether a controlled split across your users could settle it in days, and if so, run it before arguing further.
Principle 8
Build outside the monoculture
On June 7, 2010, von Ahn published his case for building in Pittsburgh. The talent at CMU was world-class, and the competition to hire it was a fraction of what it was in California. The intellectual environment was less homogeneous: in Silicon Valley, he wrote, "everybody thinks exactly the same as each other (most even dress the same as each other)." reCAPTCHA never left Pittsburgh after Google bought it. Duolingo became the city's first billion-dollar tech startup.
Distance from the center is a cost and a hedge. It costs proximity to capital and networks; von Ahn solved that by raising from Union Square Ventures, NEA, Kleiner, and Google without moving. It hedges against herd thinking and talent wars. A company that cannot be poached as easily, and does not think like its neighbors, has two advantages that compound over time.
≠
Conventional Wisdom vs. the von Ahn Approach
Where his choices diverged from startup orthodoxy.
| Conventional wisdom | Von Ahn's approach |
|---|---|
| Move to Silicon Valley to be near capital. | Stay in Pittsburgh; fly the capital in. |
| Monetize the breakthrough. | Give CAPTCHA away; monetize what it made possible. |
| Ask users what they want. | Watch what they already do and attach work to it. |
| Compete with others in your category. | Compete with social media for attention. |
| Maximize revenue per user. | Keep the content free; let the paying minority fund everyone else. |
| Control the brand voice tightly. | Let the owl become a menace. |
Tactic: Treat your location as a talent strategy: build where the best people you need are under-recruited, and bring capital and customers to you rather than going to them.
Principle 9
Stay deliberately under-monetized
Duolingo's most unusual financial choice is what it has refused to charge for. "Most of them, in fact, 93% of them, have not paid us a single cent," von Ahn told the Hall of Fame, as a point of pride. In 2022 the company reported that about 6.8 percent of monthly active users paid. In conversation with Eric Ries, he has argued for staying under-monetized and described trust with users as something like a cultural bank account, built up through restraint and spent during controversy. He has said he wants to build a hundred-year company, not satisfy quarterly expectations.
The AI backlash of 2025 tested that account. Users threatened to abandon the app; the financial results showed little damage. Restraint during the good years had bought tolerance during the bad ones. Von Ahn's own assessment, that the effect was not "very material," may have sounded tone-deaf, but it was also evidence that the savings had been made.
Tactic: Identify the features you could charge for but shouldn't, keep them free on purpose, and treat the resulting goodwill as a reserve you will eventually need to spend.
Principle 10
Name the skill you lack, then rent it from someone who has it
Von Ahn is candid about his gaps. He micromanaged his first thirty employees "the hell out of" them. He wishes he had worked on his social skills earlier. He wishes he had studied management. Most remarkably, he told his lead late-stage investor at the outset: "I'm going to be a first-time CEO. My sense is you're going to see things based on your experience that I haven't seen yet." When Henry Ellenbogen took him to dinner in 2022 and told him the market now demanded profitability, von Ahn listened, and later credited the advice with roughly tripling the company's value.
The point is not humility as a virtue. It is humility as information flow. A founder who has named his blind spots in advance gives his advisors permission to point at them. Von Ahn chose an investor whose skill, pattern recognition across companies, was exactly what a first-time CEO with an academic background lacked.
Tactic: Tell your most experienced investor or advisor, explicitly and early, which kinds of judgment you expect to be missing, so they know where to speak up.
Principle 11
Hold the product line; never skip the context
Von Ahn's two most controversial decisions show the difference between substance and communication. When Duolingo replaced its lesson tree with a linear path in 2022, users protested, he refused to reverse it, and the company kept growing. When he announced the AI-first strategy in April 2025, the substance (fewer contractors, more automation, AI-built courses) was not radically different from what Duolingo had done in January 2024. But the framing, especially the line about accepting "occasional small hits on quality," set off a backlash that forced the company to wipe its social channels. "This was on me," he said later. "I did not give enough context."
The lesson is not to avoid hard decisions. It is that a decision announced without its reasoning gets filled in by the audience's worst fears, and AI and labor are subjects where those fears are already waiting. His own contradictions, saying AI can teach anything and also that teachers can't be replaced because people need inspiring, made the gap wider.
Tactic: Before announcing any change that touches jobs or quality, write down the three fears your audience will project onto it and address each one in the announcement itself.
Principle 12
Let the brand misbehave
Duolingo's owl became famous because the internet made him a villain, and the company had the nerve to agree. The memes cast Duo as a stalker, so Duolingo made "Duolingo Push," with a man in an owl suit following users until they practiced. It killed him with a Cybertruck in February 2025 and made users resurrect him with 50 billion experience points. Von Ahn has said he had doubts about the campaign and backed it anyway, and he credits a young hire for the social strategy. Until 2019, Duolingo grew on word of mouth alone, without advertising.
For a founder whose instincts are analytical, giving up control of the brand voice is counterintuitive. It worked because the joke was true. The notifications really are relentless, and the owl's menace was a confession of Duolingo's own engagement strategy. Self-awareness is disarming. The 2025 backlash showed the limit: a brand built on irony is hard to use for corporate seriousness. The audience that loves an unhinged owl will not accept a memo in the same tone of voice.
Tactic: Find the criticism of your product that users joke about most, and consider whether the company can own the joke instead of fighting it, as long as the joke is true.
For readers who want the gentler, more inspirational version of the arc, there is Luis von Ahn: Innovating Education and Technology for a More Accessible World. The mechanisms above are what that story looks like when you take it apart.
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
— Luis von Ahn, Duolingo values letter, April 15, 2020I remember thinking it would be pretty cool if one million people ever used our website.
— Luis von Ahn, to CMU News on receiving the MacArthur Fellowship, 2006Had I won the lottery, I probably would have bought a bigger house. But I think this money is meant for something more than that.
— Luis von Ahn, Reddit AMA, 2013We spend a lot of effort making Duolingo addictive — we think the hardest part about learning anything on your own is motivation.
— Luis von Ahn, to the National Inventors Hall of FameEarly on, what motivated me was surprising, non-obvious, clever ideas. These days, I'm a lot more motivated by large-scale impact.
— Luis von Ahn, The Tim Ferriss Show, 2022Whenever somebody asks you to do something a long time from now, change the framing and say, 'What if it was tomorrow? Would you want to do this or not?'
Maxims
- Audit wasted effort first. The most valuable resource in your system may be attention people are already spending for nothing.
- Make honesty the winning strategy. If the only way to score is to tell the truth, you won't need to police the players much.
- Free is a distribution strategy, not a price. Give away what generates data and reach; charge for what that reach makes possible.
- The mission is a constraint, not a model. Keep the purpose fixed and let the revenue engine be rewritten as often as the evidence requires.
- Charge at the moment of payoff. People pay most willingly when they can see the return, such as an admission or a visa.
- Your competitor is whatever they open instead. Benchmark engagement against the most compelling app on the phone, not the nearest rival in your category.
- Run the experiment before the argument. At scale, most product debates can be settled in days by splitting users and measuring.
- Distance from the herd is an asset. Build where talent is under-recruited and thinking is less uniform.
- Name your blind spots out loud. Tell the people with more experience exactly where you expect to be wrong.
- Never announce a hard decision without its reasons. An audience left without context fills the silence with its worst fears.
