Hello, world, again

In its 2025 season Major League Hacking, which runs student hackathons, counted more than 150,000 participants across 184 events and programmes, and over 12,500 projects.^{3Major League HackingWelcome to the 2026 hackathon season8 Jul 2025} That is a project for every twelve or so hackers, which in the genre of weekend software is a respectable hit rate. It is also roughly where the useful numbers stop. Students building apps with AI have become a staple of the technology press, but the data that describe them measure enthusiasm. None of them measures income.

Start with the enthusiasm, which is real. In the Stack Overflow developer survey for 2025, 73.3% of people still learning to code used AI tools, against 78.5% of all respondents.^{1Stack Overflow2025 Stack Overflow Developer Survey: AI29 Jul 2025} MLH’s own census of hackers found that 75% of verified respondents use AI to learn technical skills, and the group concluded that “AI is now the most common way that new developers learn.”^{2Major League HackingThe data is in: AI is how developers learn now24 Aug 2026} The same post notes that only 3% of those learners rely on AI alone; most still watch videos, take courses and turn up to events.^{2Major League HackingThe data is in: AI is how developers learn now24 Aug 2026} The student with a chatbot is not replacing the classroom so much as adding a very patient tutor who never sleeps and occasionally lies.

Learning with a machine. Share using AI tools, %. Different surveys, different populations

	Using AI
All developers	78.5
People learning to code	73.3
MLH hackers learning	75

Source: Stack Overflow Developer Survey 2025; Major League Hacking¹²

The lying matters. In the Stack Overflow survey, 46% of developers distrusted the accuracy of AI tools and 33% trusted them.^{1Stack Overflow2025 Stack Overflow Developer Survey: AI29 Jul 2025} Programming has a convenient property here: the code either runs or it does not. That makes it a better subject for learning by machine than, say, history essays, and goes some way to explaining why a student can ship something in a weekend that once took a term.

Apple offers a tidy specimen. Its Swift Student Challenge named 350 winners from 37 countries and regions this year, and Apple said they “harness the power of Apple platforms, Swift, and AI tools”.^{4Apple NewsroomAI meets accessibility in this year's Swift Student Challenge7 May 2026} One winner, Karen-Happuch Peprah Henneh, said that “Something that would have taken me months to do was able to be done in three or four days.”^{4Apple NewsroomAI meets accessibility in this year's Swift Student Challenge7 May 2026} Months into days is a fair description of what these tools do. It also describes what happens to the scarcity of the output.

That is the first problem with the business case. When anyone can build a plausible app in a weekend, the app is not the scarce thing. Distribution, trust and a reason for strangers to open it twice are the scarce things, and a hackathon prize supplies none of them. We found no survey of the earnings of student-built apps, and neither Apple nor MLH reports how many projects were ever sold or even released. Pew’s survey of American teenagers and their parents, fielded last autumn, reports on schoolwork and cheating, not on building.^{5Pew Research CenterHow Teens Use and View AI24 Feb 2026} For all the talk of teenage founders, the published evidence is of teenage makers.

The best counter-argument is that income is the wrong yardstick. Nobody judges a school orchestra by its record sales. A project that teaches a 16-year-old to scope a problem, ship it and watch strangers break it is worth having whether or not it earns a cent. We agree, and note that this is an argument for treating the projects as education, which is not how they are usually pitched.

As a signal to admissions officers, they are worth less than their authors hope. A prize says that someone finished something. An AI that helps people finish things lowers the price of that signal, as cheaper printing once did for the handwritten testimonial. Admissions offices have learned to discount the bedroom charity and the curated summer programme. They will learn to discount the weekend app too, and ask what the applicant can explain without the chatbot at their elbow.

The skill that survives is the one no survey counts: knowing what to build, and noticing when the machine has built it wrongly. Students who have it should keep going. Those with eleven apps and no users have, at least, learned what a market is.