The Teenage App Boom, and the Arithmetic Underneath It

The product is a gesture everyone already makes. You hold a phone over a plate, the way people have photographed lunch for a decade, and instead of a filter you get a number: calories, protein, fat, carbohydrate. Cal AI, the app that turned that gesture into a business, launched in May 2024. Its two best-known co-founders, Zach Yadegari and Henry Langmack, were 18 and still in high school when TechCrunch first wrote about them in March 2025. By March 2026, MyFitnessPal, the incumbent in the category, had bought it, saying Cal AI had passed 15 million downloads and more than $30 million in annual revenue in under two years.

That sequence, from school corridor to acquisition in under two years, has become the founding myth of a small genre: the teenage AI app. The myth holds that the tools have collapsed the cost of building software to near zero, that the App Store will hand distribution to anyone with a good TikTok, and that a seventeen-year-old with a subscription paywall can out-earn a graduate with a law degree. The myth is not wrong about the first part. It is badly incomplete about the rest.

The cost of the first version

Start with what is true. Building a credible consumer app used to require a team: an iOS developer, a back-end engineer, a designer, and someone to make the thing reliable enough to submit. Cal AI’s first version needed none of that infrastructure in the old sense. The company told TechCrunch that the app uses models from Anthropic and OpenAI, along with retrieval techniques, and draws on open-source food and calorie image databases. The hard part, recognising a bowl of ramen and estimating what is in it, was rented, not built.

The same pattern shows up further down the age ladder. The Wall Street Journal’s reporting on teenage founders, as summarised by VnExpress, featured Nick Dobroshinsky, a 15-year-old who runs an AI financial-research site and says he has written only about ten lines of its code himself, leaning on Claude, ChatGPT and Gemini for the architecture. He attends school until 2:55 in the afternoon. His marketing department is a set of Reddit bots.

Multiply that by every bedroom with a laptop and you get a supply shock. Appfigures counted 557,000 new app submissions to Apple’s App Store in 2025, up 24 per cent from 448,000 in 2024, though still well short of the 2016 peak of more than a million. The firm’s explanation names AI tools and “vibe coding” first, followed by viral growth through TikTok and Instagram and the steady income that subscriptions make possible. The surge was large enough to strain the gate. The Next Web reported that app review times reportedly stretched from about 24 hours to as many as 30 days, and that Apple removed Anything, an app for building apps, on March 30, 2026, citing a guideline that forbids software from changing its own functionality after review. The company that built the shop is now, in effect, rationing shelf space.

This is the first complication in the myth. When the cost of making something falls toward zero, the scarce input moves. For the teenage app, the scarce input is no longer code. It is attention.

Distribution is the product

Read the Cal AI story closely and the code is the least remarkable part of it. Yadegari had published his first App Store game at 12, and in high school built a gaming site, Totally Science, which he sold at 16 for $100,000. He was not a novice whom AI turned into a programmer; he was an experienced operator whom AI made faster. More telling is the cap table. TechCrunch’s first report listed two older co-founders: Blake Anderson, 24, who had previously built ChatGPT dating-advice apps, and Jake Castillo, 28, the chief operating officer, who ran influencer marketing. The idea took shape in a San Francisco hacker house.

Fortune put the growth mechanism plainly. The team made TikToks and partnered with influencers, which Yadegari described as producing “pretty explosive growth.” That is the actual playbook of this cohort, and it is closer to the direct-to-consumer brands of the 2010s than to the garage software of the 1970s: a cheap, legible product, a paywall shown early, and a paid creator machine pointed at a demographic that lives on short video. The venture investor Kulveer Taggar made the point in the Journal’s reporting, as VnExpress relayed it: founder ages have fallen as AI tools spread, but distribution remains a major challenge.

Seen this way, AI did not democratise app-making so much as relocate its moat. The barrier used to be engineering talent, which money could hire. It is now audience-building and paid acquisition, which money can also buy, and which rewards people who already know how the feed works. A teenager who has spent ten thousand hours on TikTok has a genuine edge there. So does a 28-year-old who runs influencer campaigns for a living.

The arithmetic of a paywall

Then there is the money, which is less simple than a headline revenue figure suggests. Cal AI is free to download and costs $2.49 a month or $29.99 a year. At the annual list price, $30 million in revenue implies something on the order of a million paid subscription-years, sustained against everything that pulls subscribers away.

Before any of that reaches the founders, the platform takes its share. Apple’s Small Business Program charges a reduced 15 per cent commission to developers who earned up to $1 million in proceeds the previous calendar year; once a developer passes that threshold in the current year, the standard rate applies to future sales, which Apple’s original announcement put at 30 per cent. A breakout app clears the small-business tier quickly, so its most successful months are also its most heavily taxed.

The next cost is the one that distinguishes this generation of apps from the last. A conventional calorie counter that looks up foods in a database has trivial marginal costs; one more user is a few more rows queried. An app that sends every meal photograph to a frontier model pays for each inference. The more engaged the user, the more photographs, and the higher the bill. This is an inversion of classic software economics, in which the best customers are the cheapest to serve. Neither Fortune nor TechCrunch reported Cal AI’s margins, and this piece will not guess at them. But the structure is clear: revenue is fixed per subscriber, cost scales with use, and the provider setting the price of that use is a company with its own margin targets.

Survivorship, in numbers

The deeper problem is that the Cal AI story is a single observation drawn from the far end of a distribution, and the distribution is brutal. RevenueCat, whose tools are used by more than 75,000 app developers, published the numbers in its 2025 report: “At $8,880 in earnings, the top 5% of newly launched apps make over 400x as much revenue after two years as the bottom 25%, who make no more than $19.” Read that sentence slowly. The threshold for the top twentieth of new subscription apps, after two full years, is less than $9,000. Cal AI is not in the top 5 per cent. It is in the top sliver of the top 5 per cent, and the median teenager who ships an app this year should expect to earn roughly what a summer job pays, or less.

Retention compounds the problem. The same report found that “Nearly 30% of annual subscriptions are canceled in the first month.” And AI apps, the category this cohort is crowding into, behave worse than average over time. RevenueCat’s 2026 data, reported by TechCrunch, put annual-plan retention at 21.1 per cent for AI apps against 30.7 per cent for others, with AI apps’ annual subscriptions cancelled 30 per cent faster at the median. The same data contain the explanation for why so many people build them anyway: AI apps converted trials at 8.5 per cent against 5.6 per cent, and earned a median realised lifetime value on annual plans of $30.16 against $21.37. They are, in short, good at the first sale and bad at the second year. That is the profile of a novelty, and novelties are what short-video marketing sells best.

Cal AI’s own early numbers came with a caveat that should be printed on every revenue screenshot. In March 2025 Yadegari told TechCrunch the app had more than 5 million downloads, retention above 30 per cent and more than $2 million in revenue the previous month; TechCrunch noted that it could not validate the download and revenue claims. The acquirer’s later figures lend them weight in retrospect. Most teenage founders posting monthly-recurring-revenue charts will never have an acquirer to vouch for them.

What breaks when building is cheap

The skeptics have a second argument, and it is not about money. When the cost of shipping falls, so does the cost of shipping carelessly. Quittr, an app for people trying to stop watching pornography, was built by what 404 Media described as “a couple of 20-year-old developers” and claimed to make $500,000 a month. In March 2026 the outlet reported that it had exposed sensitive data on hundreds of thousands of users, many of them minors, including their ages, how often they masturbate and how pornography makes them feel. 404 Media had first reported the exposure in January but could not name the app, because the developers had not fixed it.

The lesson is not that young founders are uniquely reckless; adult-run companies leak data constantly. It is that AI tools compress the part of software that is visible, the interface and the features, while doing little for the parts that are not: security review, data handling, the dull discipline that a senior engineer would have insisted on. An app can now reach a million users before anyone involved has thought hard about what happens to their information. The users, in this case, included children.

The steelman, and the turn

None of this makes the boom illusory. A real thing has happened: the minimum viable team for a consumer software business has shrunk to two or three people, and some of them can be teenagers. Cal AI reportedly grew from eight full-time employees in early 2025 to more than 30 people that autumn, and MyFitnessPal’s chief executive, Mike Fisher, said of the sale that the team “didn’t have to sell.” Those are the facts of a serious business, not a fluke.

And the standard path did not exactly compete. Yadegari, by TechCrunch’s account, was rejected by 15 of the 18 top colleges he applied to despite a 4.0 grade-point average, a fact he shared on X. He enrolled at the University of Miami anyway, and told Fortune he regards it as “$100,000 a year for a vacation.” It is a good line, and a revealing one: the person best placed to declare college obsolete is still paying for it, as a hedge, a social life, or both.

That ambivalence is the most honest summary of the teenage app boom. AI has made the first version of a product almost free, which means the first version is no longer the achievement. What remains scarce is what has always been scarce in consumer markets: an audience, the stamina to keep subscribers past the first month, the margin left after the platform and the model provider are paid, and the judgment not to leak your users’ secrets. Those are not things a high schooler learns from a chatbot. They are, inconveniently, a lot like the things an entry-level job was supposed to teach.