The Streak
On the night of August 23, 2022, a 50-year-old Dutch IT consultant named Tobi Fondse sat down after dinner, pulled out his phone, and opened an app with a green owl on it. He and his wife, Marisa, had completed at least one Duolingo lesson every day for more than 400 consecutive days — not because anyone was making them, not because a boss required it, not because a degree hung in the balance, but because they didn't want to break the streak. Fondse wanted to order croissants au jambon in France without pointing. This is the smallest possible unit of ambition. And yet this moment — a middle-aged man on a couch in the Netherlands, tapping through French conjugations at 9 PM because a cartoon owl would cry if he didn't — contains the entire architecture of a company now worth, at various points in its young public life, between $8 billion and $17 billion. The streak is the product. The streak is the business model. The streak is the moat.
Duolingo is a company that makes approximately $748 million per year teaching people things they mostly don't need to know, using methods borrowed from slot machines and mobile games, distributed for free on phones that cost less than a single semester of community college language instruction, and yet — here is the part that matters — it works. Not perfectly. Not to fluency. But measurably, consistently, at a scale no language program in human history has achieved. By the end of 2024, more than 40 million people were using Duolingo every single day, a figure that had grown 51% year-over-year and roughly 10x since 2019. The company had 113 million monthly active users, 10.3 million paid subscribers, and a DAU-to-MAU ratio that would make most social media companies envious. It was profitable. It was generating free cash flow at a 30%+ margin. And its founder — a Guatemalan computer scientist who had previously invented the technology that made you prove to websites that you were not a robot — was declaring that the company was "at 1% of what we could achieve."
That claim sounds like the kind of thing CEOs say. But it also captures something real about the strangeness of Duolingo's position: a consumer internet company wearing the disguise of an education company, or perhaps an education company that has internalized the operating logic of consumer internet so completely that the disguise has become the thing itself.
By the Numbers
The Duolingo Machine
$748MFY2024 revenue
40.5MDaily active users (Q1 2025)
116.7MMonthly active users (Q1 2025)
10.3MPaid subscribers (FY2024)
~$15BMarket capitalization (late 2024)
40%+Revenue CAGR since IPO
30%+Free cash flow margin
800+Employees worldwide
The Human Computer
Luis von Ahn was born in Guatemala City in 1978, the son of a doctor, raised by his mother and grandmother in a country where 55% of the population lived in poverty and the quality of your education was a direct function of your parents' bank account. His mother spent her entire salary on his schooling — a private English-language education that gave him access to a world most Guatemalans couldn't reach. When he was eight, she bought him a Commodore 64. He wanted a Nintendo. The computer, he later acknowledged, changed his life.
Von Ahn went to Duke for mathematics, then Carnegie Mellon for a PhD in computer science, and there — in the fall of 2000, as the dot-com bubble was deflating — he attended a talk about ten problems Yahoo couldn't solve. One problem caught him: bots were registering millions of fake email accounts, and the company had no way to distinguish them from humans. Von Ahn, inspired by a passage in a Douglas Hofstadter book about computers' inability to recognize distorted text, built a solution with his adviser Manuel Blum. They called it CAPTCHA — Completely Automated Public Turing test to tell Computers and Humans Apart — and gave the code to Yahoo for free. Within three years, nearly every large internet company was using some version of it. CAPTCHA did not make von Ahn rich. It made him mildly infamous. "Oh, you came up with that? I hate you," people would tell him.
What it also did was establish the intellectual pattern that would define his career: the idea that vast quantities of human attention, distributed across millions of tiny tasks, could be channeled toward something productive. This was the animating insight of his entire body of work. After CAPTCHA, he created the ESP Game, which paired random players online and had them label images — ten million people played, and their collective judgments helped Google improve image search. Then came reCAPTCHA, the leap that made his reputation. By 2007, people were solving CAPTCHAs 200 million times a day, spending roughly ten seconds each time — collectively, 500,000 hours of cognitive labor per day, wasted on proving humanity to machines. What if, von Ahn wondered during a drive from Washington to Pittsburgh, each of those ten-second micro-tasks also contributed to digitizing a book? He built a system that took words optical character recognition software couldn't read from scanned books and embedded them in CAPTCHA challenges. Users proved they were human and transcribed the New York Times archive, simultaneously. He sold reCAPTCHA to Google in 2009 for a reported $30 million and won a MacArthur "genius" grant at 28.
The through-line is worth pausing on. CAPTCHA, the ESP Game, reCAPTCHA — each one took a pool of distributed human attention and turned it into something more valuable than the individual contributor realized. Von Ahn coined the term "human computation" for this. What he was really building was an intellectual framework for understanding how millions of people, each spending a few seconds, could collectively accomplish tasks that neither humans nor computers could manage alone. Duolingo would be the application of that framework to education. But unlike reCAPTCHA, where the user's contribution was invisible to them, Duolingo would need the user to be the primary beneficiary. The human computation, this time, would be on yourself.
The Translation Machine That Became a School
The original idea for Duolingo, circa 2009, was not the company it became. Von Ahn and his PhD student Severin Hacker — a Swiss-born computer scientist whose surname, improbably, actually was Hacker — wanted to build a system that would teach people languages while simultaneously harnessing their translations for commercial purposes. The business model was elegant in theory: companies like CNN or BuzzFeed would pay to have their websites translated into other languages, and the translations would be generated, collaboratively and incrementally, by language learners on the platform. Learning was the product; translation was the revenue engine.
Hacker, who held a BS from ETH Zurich and was pursuing his PhD under von Ahn at Carnegie Mellon, shared his adviser's conviction that education should be free. Both were non-native English speakers. Both understood, viscerally, that language acquisition was the single most reliable economic escalator available to the global poor. "In Guatemala, everyone wants to learn English to get out of poverty," von Ahn would later say. "The ironic thing is if you want to learn English, it takes $1,000" with traditional programs. The injustice wasn't abstract for him — it was autobiographical.
They launched Duolingo in 2012 with six languages: English, Spanish, French, German, Portuguese, and Italian. The translation model worked, technically. Users would learn vocabulary and grammar through exercises, and their translations of real web content — aggregated and compared across multiple users to ensure accuracy — were sold to publishers. It was crowdsourced translation bootstrapped onto a free language course.
Bill Gates called Duolingo his favorite education app. In 2013, Apple named it App of the Year. More than 100,000 new users signed up every day.
But the translation business model had a ceiling. The quality was inconsistent. The volume of content needing translation was finite. And the real magic was happening on the other side of the equation — the learning side. Users weren't coming for the privilege of translating CNN articles. They were coming because the app was, as von Ahn put it, "like Candy Crush but you learn something." The gamification — experience points, virtual currency called "Lingots," a green owl mascot that cried when you missed a lesson — was working better than anyone expected. Users were forming habits. They were completing lessons in grocery store lines, in doctors' waiting rooms, on buses. The format — three minutes here, five minutes there, always on your phone — was uniquely suited to how people actually learned, which is to say: in fragments, opportunistically, and only if it doesn't feel like work.
Duolingo pivoted away from the translation model and toward a freemium subscription business, supported by advertising for free users and premium subscriptions for those willing to pay for an ad-free experience. The mission remained the same — free education for all — but the commercial architecture shifted fundamentally. The "human computation" component receded into the background. What remained was a machine designed to do one thing exceptionally well: keep you coming back.
The main way in which we grow is by word of mouth, so people tell their friends.
The [Attention](/mental-models/attention) Factory
To understand what Duolingo actually built, you need to understand what it is competing against. Not Rosetta Stone. Not Babbel. Not even other education apps. Duolingo competes against every app on your phone for your most scarce resource: the moments when you might voluntarily do something that isn't entertaining.
The company understood this from the start, but the understanding deepened over time into something closer to an engineering discipline. Jorge Mazal, who served as vice president of product, described the team's approach as fundamentally data-driven: "If we want to know whether we should teach plurals before adjectives, 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." By the mid-2020s, Duolingo was running more than 750 A/B tests per quarter — roughly eight tests a day, every day, relentlessly optimizing the product.
What emerged from this optimization culture was a system with three interlocking layers. The first was pedagogical: bite-sized lessons, spaced repetition, adaptive difficulty powered by an AI algorithm called Birdbrain that tracked each user's proficiency at a granular level and served exercises calibrated to their zone of proximal development. An independent study found that 34 hours of Duolingo instruction produced learning equivalent to one university semester of language study.
The second layer was behavioral: the gamification engine that borrowed liberally from mobile gaming. Streaks — the consecutive-day counter that Fondse and his wife were so desperate to maintain — proved to be the single most powerful engagement mechanic. The concept is simple: do one lesson per day, and your streak counter increments. Miss a day, and it resets to zero. Loss aversion, one of the most robust findings in behavioral psychology, does the rest. Users with long streaks experience the prospect of breaking them as genuinely painful. Some set alarms. Some complete lessons at 11:58 PM. Some — and this is not a joke — have maintained streaks exceeding 3,000 days, nearly a decade of daily practice without a single missed day. The leaderboard system, which placed users in competitive "leagues" that reset weekly, added a social competitive layer. Hearts — limited "lives" that depleted with wrong answers, replenished by watching ads or subscribing — created a scarcity mechanic that drove both engagement and monetization.
The third layer was emotional: Duo the owl. The mascot began as a logo — "basically a chicken nugget with a face," as head of art Greg Hartman described it. Its green color was an inside joke; co-founder Severin Hacker had strong feelings about not using green in the branding, so von Ahn made the mascot green specifically to annoy him. Over time, Duo evolved into a full character with a personality — encouraging when you practiced, devastated when you didn't, and occasionally, in the company's social media presence, genuinely unhinged. The owl would appear in push notifications with escalating emotional intensity. "Looks like you forgot your Spanish lesson," read the mild version. The aggressive ones — memed relentlessly by users — implied that Duo would show up at your house.
This wasn't accidental. The size of the puddle of Duo's tears when a user failed a task was researched to optimize return rates. Every pixel of emotion was A/B tested.
From logo to cultural phenomenon
2012Duo debuts as a simple green owl logo — no facial expression, purely functional.
2014Duo receives expressions, including tears that appear when users fail tasks. Tear puddle size is A/B tested.
2017Push notification strategy intensifies; "passive-aggressive owl" memes begin circulating on social media.
2020Duolingo's TikTok account launches; Duo becomes an "unhinged" cultural character, generating hundreds of millions of views.
2025Duo is "killed" in a two-week marketing stunt; users mourn globally. The resurrection drives massive engagement.
The Social Media Company Disguised as an Education Company
In 2020, Duolingo hired Zaria Parvez as a social media manager. What followed was one of the most improbable brand-building campaigns in modern marketing — not because it was expensive or sophisticated, but because it was cheap, weird, and relentlessly committed to making the green owl a character in popular culture.
The Duolingo TikTok account became a phenomenon. Duo appeared in a full-body mascot costume, lurking in the Duolingo offices, participating in TikTok trends, making jokes that had nothing to do with language learning, flirting with other brand accounts, and generally behaving like an entity that had escaped corporate control. The tone was described, accurately, as "unhinged." A video of Duo ice skating for April Fools' Day generated roughly 26 million views. The account accumulated hundreds of millions of views per month at its peak. In 2024, Ad Age named Duolingo one of its Marketers of the Year.
The genius of this strategy was that it cost almost nothing in paid media spend. Duolingo's marketing budget was minuscule relative to competitors; the company spent essentially nothing on paid user acquisition. Instead, it relied on organic virality — the TikTok account, word-of-mouth recommendations (Bill Gates mentioned Duolingo on Reddit in 2015, and the company still cites it), and the inherent shareability of streaks and leaderboard achievements. When someone hits a 365-day streak, they tell their friends. When someone sees a meme of Duo threatening bodily harm for missed lessons, they download the app. The owl was doing the marketing work that other consumer companies spent hundreds of millions of dollars on — not because Duolingo had discovered some secret, but because it had built a product that was inherently viral and then given that virality a mascot with personality.
This is the paradox at the center of Duolingo's brand strategy: the company's identity is simultaneously educational and absurd, mission-driven and trollish, earnest and ironic. Von Ahn has spoken openly about how the product's playful aesthetic causes some observers to underestimate its sophistication. "People confuse fun with a lack of rigor," he has noted. But the underestimation is itself a competitive advantage — it made the brand approachable in a way that traditional education brands, with their sterile authority, never managed.
Zero to IPO in a City That Wasn't Silicon Valley
Duolingo's financial trajectory from founding to IPO is a study in patient capital and deliberate restraint. The company raised approximately $183 million across multiple venture rounds from investors including Union Square Ventures, Kleiner Perkins, New Enterprise Associates, and Alphabet's CapitalG. Ashton Kutcher and Tim Ferriss participated in early rounds. Valuations climbed from roughly $45 million in a 2012 Series A to $700 million by 2018 and $2.4 billion by a 2020 funding round that reflected the COVID-19 surge in language learning activity.
The pandemic was an accelerant. Over 30 million new users started learning a language on Duolingo in the weeks following global lockdown measures. People trapped at home, looking for something to do that felt productive, turned to the app in enormous numbers. Monthly active users surged. Engagement deepened. The business, which had generated zero revenue for its first five years of existence and only $13 million in 2017, hit $161.7 million in revenue in 2020 — a 129% increase year-over-year.
📈
From Pittsburgh to NASDAQ
Key milestones in Duolingo's journey to the public markets
2011Von Ahn and Hacker co-found Duolingo at Carnegie Mellon. Early team works above an Italian restaurant in Pittsburgh.
2012App launches with six languages. Over 100,000 users sign up daily within months.
2013Apple names Duolingo App of the Year. Google names it Best of the Best.
2017Revenue reaches $13 million — after five years of generating zero revenue. Revenue grows 1,300% year-over-year.
2020COVID-19 drives 30 million new users. Revenue hits $161.7 million.
2021IPO on NASDAQ at $102 per share; market cap reaches $3.7 billion on first day of trading.
Duolingo went public on July 28, 2021, listing on the NASDAQ under the ticker DUOL at an initial offering price of $102 per share. The stock opened at $141.40 — nearly 39% above the IPO price — giving the company a valuation of approximately $3.7 billion. It was the largest tech IPO to come out of Pittsburgh. Von Ahn rang the bell, and the company's S-1 filing laid out its strategic logic with unusual clarity: the freemium model was not charity, it was a growth engine. More free users meant more data, more data meant better personalization, better personalization meant higher engagement, higher engagement meant more conversions to paid subscribers, and more subscribers meant the revenue to keep everything free. It was a flywheel, and by IPO, it was spinning fast.
The filing also revealed something the market would grapple with for years: the company's extraordinary dependence on app store distribution. In 2020, 51% of Duolingo's revenue came through Apple's App Store and 19% from Google Play. The 30% commission Apple charged on in-app purchases was, in effect, a toll that ate directly into the company's economics. It was the single largest cost item in the business model of a company that sold primarily through mobile subscriptions.
The Monetization Riddle
Duolingo's business model is a bet on a specific theory of human behavior: that if you give something away for free and make it good enough, a sufficient percentage of users will voluntarily pay for a better version. This is the freemium model, practiced across consumer software, but Duolingo's version has some unusual properties.
The free product is genuinely complete. You can learn any of 42 languages — from Spanish to Klingon to Scottish Gaelic — without paying a cent. The ads are interstitial, appearing between lessons, and while annoying, they don't degrade the learning experience. The heart system creates friction: free users have limited "lives," and each wrong answer costs one. Run out of hearts, and you wait — or watch an ad to earn one back. This is the primary conversion lever. Paying subscribers get unlimited hearts, no ads, offline access, and progress analytics.
The subscription tiers evolved over time. "Super Duolingo" (formerly Duolingo Plus) offers the core premium experience. "Duolingo Max," launched in 2023, adds AI-powered features built on GPT-4: a "Roleplay" feature for practicing conversations with AI characters, a "Video Call" feature for real-time speaking practice with the character Lily, and "Explain My Answer," which provides detailed explanations of why an answer was right or wrong. Max is priced significantly higher — roughly double the Super tier — and by Q2 2025, Max subscribers accounted for 8% of the paid subscriber base, up from 5% a few quarters earlier. A family plan, which allows up to six users under a single subscription, became a meaningful growth driver as household adoption increased.
The conversion math is revealing. Of Duolingo's roughly 116 million monthly active users as of early 2025, approximately 9.5 million were paid subscribers — a conversion rate of about 8-9%. This is low compared to some consumer subscription businesses, but the absolute numbers are enormous and growing. Paid subscribers grew 47-54% year-over-year through most of 2024. More importantly, the company was expanding revenue per subscriber through tier upgrades and family plan adoption while simultaneously growing the free user base that fed the conversion funnel.
Advertising revenue, while a smaller portion of total revenue, scales directly with daily active users. The Duolingo English Test (DET) — an affordable, remotely proctored English proficiency exam accepted by over 5,000 universities and institutions — represents a third revenue stream, though it remains a modest contributor.
We believe that we are still in the early stages of our monetization journey, and discovering multiple avenues to enhance bookings.
The financial results told the story. Full-year 2024 revenue reached $748 million, up approximately 40% year-over-year. Subscription revenue accounted for roughly 76% of total revenue. Adjusted EBITDA margins expanded from the low teens to the mid-20s over the course of 2024.
Free cash flow margins exceeded 30%. The company was sitting on more than $1.1 billion in net cash with no debt. For a company that generated zero revenue for its first five years and $13 million as recently as 2017, the trajectory was extraordinary — and it was accelerating.
The AI Moment
In early 2023, when OpenAI released GPT-4, most education companies experienced it as a threat. Duolingo experienced it as an opportunity — and moved faster than almost anyone expected.
Within months of GPT-4's launch, Duolingo announced Duolingo Max, the AI-powered subscription tier that used OpenAI's technology to create conversational practice, detailed answer explanations, and eventually real-time video calls with AI characters. The features represented a qualitative leap in what a language-learning app could offer. For the first time, users could practice speaking — the skill that was hardest to develop through an app — with an AI tutor that responded naturally, corrected errors, and adapted to the learner's level. "If you look at what is the best way to learn math, music, golfing, doesn't matter what, it is with a one-on-one tutor," CTO Severin Hacker explained. "The promise of technology and A.I. is that you can reach that level of efficacy but actually make it accessible to everyone."
The AI integration went deeper than the consumer-facing features. On the content production side, generative AI transformed Duolingo's operational economics. The company reported that AI had increased content production velocity by roughly 18x since 2021 — from 425 content units published in 2021 to 7,500 in 2024. Courses that once took teams of linguists and content designers months to build could now be created in a fraction of the time. This wasn't merely an efficiency gain; it was a structural change in the company's cost curve and its ability to expand into new subjects and languages.
But the AI moment also brought a reckoning. In January 2025, von Ahn sent an internal memo declaring Duolingo's new "AI-first" strategy, which included the announcement that the company would gradually phase out the use of contract workers for content creation as AI took over those tasks. The memo leaked. The backlash was immediate and fierce. Headlines screamed about AI replacing human workers. The nuance — that Duolingo had been reducing contract workers for years as AI capabilities improved, and that full-time headcount was actually growing — was largely lost. Von Ahn later acknowledged the communications misstep publicly: "I did not expect the blowback."
The episode crystallized a tension that would define Duolingo's next era. AI made the product dramatically better and cheaper to operate. It opened up new subjects — math, music, and eventually chess — that would have been prohibitively expensive to build with purely human content teams. It enabled the "Video Call" feature that drove Max subscription adoption. But it also invited the question that every Duolingo investor now grapples with: if AI is good enough to teach languages, is it also good enough to make language learning unnecessary?
The Existential Question
In August 2025, Google launched new AI-powered features for Google Translate that included custom language practice lessons — conversational practice, progress tracking, and personalized learning goals, all built on Gemini. Duolingo shares fell 3% on the news. Apple, at its June 2025 WWDC, announced live-translation features for iMessage, FaceTime, and phone calls. ChatGPT could already conduct fluent conversations in dozens of languages, serving as a free, infinitely patient conversation partner.
The bear case writes itself. If real-time translation becomes good enough — and it's getting better at a rate measured in months, not years — then the motivation to learn a language diminishes. Why spend 400 consecutive days studying French when your AirPods can translate in real-time? Why pay $167 per year for Duolingo Max when ChatGPT can have a conversation with you in French for free? Only about 10% of Duolingo's users are paying customers, von Ahn acknowledged to the New York Times. The other 90% might be the first to wonder whether the whole exercise is worth it.
Von Ahn's response was characteristically direct: "Just having conversations in French on something like ChatGPT gets pretty boring after a while. It doesn't keep you there. We keep you on task with all the gamification." This is not a technical argument. It's a behavioral one. And it may be the most important strategic insight in the company's arsenal. The question is not whether AI can teach a language — it can. The question is whether anyone will stick with it long enough to actually learn. Duolingo's entire moat is the answer to that question. The streak, the leaderboard, the crying owl, the loss aversion mechanics, the 750+ A/B tests per quarter optimizing the probability that you come back tomorrow — none of this exists in ChatGPT or Google Translate. The technology of learning is increasingly commoditized. The psychology of learning is Duolingo's defensible territory.
The stock market, however, was unconvinced. From a peak of roughly $545 per share in May 2025, Duolingo's stock fell to as low as $166 — a 70% decline — as the AI narrative shifted from tailwind to existential threat. The company's bookings growth, a leading indicator, decelerated. For the first time since 2022, bookings growth fell behind revenue growth, and the divergence was widening.
Just having conversations in French on something like ChatGPT gets pretty boring after a while. It doesn't keep you there. We keep you on task with all the gamification.
The Platform Bet
The response to the existential question was not to double down on language learning alone. It was to become something larger.
In late 2023, Duolingo launched math and music courses on the platform. The music course taught concepts like pitch, meter, and rhythm, complete with an on-screen keyboard. The math course focused on practical skills — calculating tips, understanding percentages, basic algebra. The company later added chess. These weren't random choices. They were subjects that shared key properties with language learning: they could be taught in short, interactive bursts on a phone; they benefited from daily practice; they were universally relevant; and they could leverage the same gamification engine that made language learning sticky.
CTO Hacker explained the logic: "The way you learn languages on Duolingo lends itself well to those two subjects. You can learn them on the phone. You kind of pick it up on the go." The company had considered everything from coding to history to geography before settling on subjects that mapped to its existing pedagogical format and engagement mechanics.
The strategic implications were enormous. Language learning, while large, is a defined market. The broader edtech market, estimated at $56 billion, was dramatically bigger. By expanding into multiple subjects while retaining the same gamified, freemium, mobile-first approach, Duolingo was transforming from a language app into a learning platform — what some analysts began calling "the Spotify of education." The company's investor relations page now described the mission as building "a 100-year company that transforms how the world learns everything."
This was the Hacker formulation of Duolingo's endgame: an educational ecosystem where AI enables personalized, one-on-one tutoring across arbitrary subjects, delivered through the same engagement mechanics that made language learning work. "We're at 1% of what we could achieve with this company," he said. The AI that threatened to commoditize Duolingo's core product was, in this telling, also the engine that would power its expansion into territories that no language app had reason to enter.
Whether the market believed this narrative was, as of late 2025, very much in question.
The Culture in the Machine
Duolingo published its internal handbook in February 2025 — an unusual move for a public company, and a deliberate one. The document laid out five operating principles that the company described as "not aspirational" but "lessons we've learned through experience."
Take the Long View: "If it helps in the short-term but hurts Duolingo in the long-term, it's not right."
Raise the Bar: "To change how the world learns, we must do world-class work."
Ship It!: "For a good idea to become reality, we need to push experiments with a sense of urgency."
Show Don't Tell: "We use clear, concise communication that is grounded in data and real impact."
Make It Fun: "We bring a sense of humor, joy, and imagination to everything we do."
The company maintained a greater than 90% annual employee retention rate. More than three-quarters of its approximately 800 employees worked in engineering, design, product, or research — a ratio that inverted the typical consumer company's allocation toward sales and marketing. Pittsburgh, where the company remained headquartered at 5900 Penn Avenue, offered advantages that Silicon Valley could not: lower cost of living, commuting times measured in minutes rather than hours, and the ability to be the most prominent tech company in the city rather than one of hundreds. "Being a locally-grown, scaling tech company is an advantage in a smaller city," VP of Workplace Sean Devlin noted, "because of the local impacts we can have."
The culture was, by design, quirky. Von Ahn described the company as having been "started with a few dozen nerds above a Pittsburgh sports bar." The humor that animated the TikTok account and the owl mascot was not a marketing overlay — it was the native register of the organization. This was a company where the green in the logo was an inside joke between co-founders, where the CTO's actual surname was Hacker, where the handbook was subtitled "14 years of big learnings in one little handbook," and where the operating philosophy was captured in the phrase "The Green Machine": gather excellent people, let them experiment, and double down on what works.
What made this culture functional, rather than merely charming, was the experimentation velocity it enabled. The 750+ A/B tests per quarter were not bureaucratic exercises — they were the core of how the product evolved. Each test produced data. Data informed decisions. Decisions shipped quickly. The cycle repeated, thousands of times, each iteration making the engagement engine slightly more effective, the learning slightly more efficient, the conversion funnel slightly wider.
The Guatemalan Computer Scientist and the Hundred-Year Company
There is a specific quality to Luis von Ahn's career that bears attention. Every major project he has built — CAPTCHA, reCAPTCHA, the ESP Game, Duolingo — has been structured around the same fundamental insight: that distributed human attention, properly channeled, can accomplish things that neither individuals nor machines can achieve alone. CAPTCHA turned security challenges into a test for humanity. reCAPTCHA turned that test into book digitization. The ESP Game turned image labeling into play. Duolingo turned language learning into a habit.
Each iteration was more ambitious than the last, and each captured more of the value for the user. CAPTCHA users got nothing for their trouble. reCAPTCHA users unknowingly digitized books. ESP Game players got entertainment. Duolingo users get an education. The trajectory describes a moral arc — from extraction to exchange to gift — that reflects von Ahn's own journey from a country where education was a luxury to a company that insists it be free.
The man himself remains a professor at Carnegie Mellon, where he still technically holds a faculty position. He is, by most accounts, awkward in the way of computer scientists — analytical, direct, more comfortable with data than with corporate messaging, as evidenced by the "AI-first" memo backlash. He has a tendency to say things that are true but impolitic, like acknowledging that 90% of his users don't pay, or noting that the company is "at 1% of what we could achieve" — a statement that is either visionary or delusional depending on your investment thesis.
His co-founder Hacker, who serves as CTO and sits on the board, is the quieter complement — the systems architect to von Ahn's public intellectual. Together they represent an unusual founding pair: two immigrants, both non-native English speakers, both trained in the formal rigor of Carnegie Mellon computer science, who built a product that teaches language through play and generates hundreds of millions of dollars from users who were never forced to pay for anything.
As of Q2 2025, daily active users had grown 40% year-over-year. The company was projecting approximately $991 million in revenue for full-year 2025. The AI-powered Video Call feature was driving Max subscription adoption. Math and music courses were gaining traction. The balance sheet held over $1.1 billion in cash with no debt. The stock, after its 70% decline, was beginning to stabilize as investors tried to reconcile two competing narratives: that Duolingo was a growth compounder with a decade of expansion ahead of it, or that it was a language app whose core product was being commoditized by the very technology it had embraced.
In the Duolingo handbook, published a month after von Ahn's AI-first memo ignited controversy, there is a line that captures the company's self-image with characteristic precision: "A culture like Duolingo's doesn't come from some corporate playbook — it had to be built from scratch."
The green owl on the home screen — the one that will cry if you don't practice, that will celebrate if you do, that was born as a joke between two friends and became a cultural icon worth billions — blinks. It is waiting for your lesson. It has been waiting all day.
