Why “Learn to Code” Failed

Between 2011 and 21, the number of Computer Science graduates at UC Berkeley increased by one thousand, one hundred, and six percent. If this trend continues — as one professor observes — all 30,000 of its undergrads will study Computer Science in just seven years. By 2059, the school will churn out more “Computer Scientists” than there are people in California! Meanwhile, get this: there are fewer software developers employed in the United States today than there were six years ago. Clearly, something has gone terribly wrong. And Berkeley isn’t even the most Computer Science-obsessed university. Not even close. This is UCB, this is CalTech, and this is MIT. You’re not reading this wrong. A full forty percent of the school that once helped invent the radar, spreadsheet, and lithium-ion battery studies the same one subject. Last year, a grand total of seven students graduated from MIT with a Bachelor of Science in Chemistry. Two-hundred and sixty-six majored in Computer Science and Engineering. And that’s not even its largest CS major — Electrical Engineering and Computer Science is over twice as popular. Over the past 15 years, U. S. colleges have transformed from multidisciplinary institutions of higher education to Microsoft training programs that also dabble in philosophy and physics on the side. In 2023, Berkeley joined MIT and Cornell in even creating an entire College of Computing — its first new school since Eisenhower was president. It’s not hard to see how we ended up here. The iPhone was released in 2007; Uber, AirBnB, and Instagram soon after. America’s cultural center of gravity was shifting from stuffy Wall Street board rooms to youthful Bay Area garages. Shows like The Social Network and Silicon Valley mythologized the tech founder — the 21st century’s more meritocratic, more visionary, and certainly more eccentric answer to the industrial tycoons of the past. Then, our nation’s leaders endorsed programming with their official seal of approval, elevating it from a promising new career to “the future. ” While in office, President Obama called coding a “ticket to the middle class. ” The White House celebrated Computer Science Education Week, promoted “Hour of Code,” and warned of a dire shortfall of 1 million STEM graduates by 2022. New York Mayor Michael Bloomberg even announced he, personally, would “learn to code” in 2012 (although there’s no evidence he ever did). There was something for everyone in that 3-word-mantra, “Learn to code…” For Republicans, this directive neatly aligned with their vision for more vocational, career-oriented education. For corporations, it promised a steady supply of skilled labor — I might add — that came pre-trained, at taxpayer expense. For national security hawks, coding helped strengthen America’s global power, enabling it to “out-compete” China. If diversity was more your cup of tea, one could tell a pretty convincing story about how this was really all about expanding opportunity. And for Democrats steering the economy through the Great Recession, “learn to code” sounded empowering, hopeful, and — most importantly — vague enough to absorb virtually any of the difficult questions posed by globalization at an especially fraught moment. Layoffs? Automation? Outsourcing? …Not to worry if we all just rolled up our sleeves and, say it with me: “learned to code! ” Red and blue states alike rushed to teach Python and Javascript in classrooms. Today, nearly 15,000 U. S. high schools offer Computer Science classes. Even 37% of middle schools do. Eleven states go even further: requiring all students take a computer science class to graduate. Absent any dissenting voices, there was every incentive to exaggerate the ease of coding and romanticize the life of a programmer. Obama told us “just about anyone can become” a computer scientist, quote, “with a little hard work…” Years later, President Biden echoed this sentiment in his signature no-nonsense

tone: quote “anybody who can throw coal into a furnace can learn how to program, for god’s sake. ” One popular article from 2012 — when U. S. unemployment was still well over 8% — declared, “to anyone out there who says you can’t get a job: You can have one. A fun one. Learning code is not about numbers and mathematics. It’s more like architecture…” Well, young Americans didn’t need to be told twice! Younger Millennials and older Gen Z looked around, saw their siblings struggle to find work after the Great Recession, and decided they’d take the ‘cool’ job that lets you earn six figures working from a bean bag while wearing a hoodie, requires a bachelor’s degree at most, and is apparently remarkably “quick” and “easy” to learn, please. But although Microsoft and Facebook and Apple were ready for this massive influx of programmers, universities most certainly were not. The popularity of college majors, of course, fluctuates all the time. Seasoned law school admissions officers still remember the sudden rise in applications following the hit 80s drama “L. A. Law,” just as pharmacy school briefly became popular during the early 2000s pharmacist shortage. But the Computer Science boom was something else entirely. The basic problem for schools was this: while the total number of undergraduate CS students tripled in just 10 years, the number of Ph. D. students — who become professors — has stayed more-or-less the same. The reason for this is no mystery: a Ph. D. student can expect to receive a $40,000/year stipend — if they’re lucky. That same student can easily earn $200,000 or more at Amazon or Netflix. In other words, there isn’t anyone to teach all those undergraduates. The result, in many Computer Science departments, is an impersonal, factory-like experience. Professors are perpetually stressed and invariably overworked. Class sizes are massive — 4, 5, 6 hundred students in introductory courses. Undergrads are even deputized as TAs — meaning the person who took this class last semester might be the only person you can ask questions. And yet schools still turn eager students away. It’s become common at many universities to make students re-apply to their Computer Science program at the end of their first or second year, for example. At Swarthmore, the University of Maryland, and UCSD, students are entered into a lottery! Meaning, you can apply and be accepted to Swarthmore (not to mention pay just over $90,000/year) only to later be denied access to your preferred major by random number. Needless to say, this leaves many students feeling… disillusioned with the college experience. They graduate with debt, with little career help from their school, whose resources are spread thin, and with minimal — if any — contact with their professors along the way. On top of all that, many of these students feel they were inadequately prepared for the job market. Because many CS programs have their origins in math and engineering departments, their curricula often focus more on the how — the general algorithms, ways of thinking, and processes — than the what — the specific syntax, programming languages that are currently marketable, or skills tested in modern job interviews. Graduates will often have dabbled in half a dozen languages but truly mastered none, forcing them to learn on their own. It wasn’t hard to imagine there might be a better way. And this being the era of quote “disrupting” “outdated” industries like taxis, hotels, and DVD stores, the proposed “solution” was wrapped in lofty, emancipatory language: education was about to be revolutionized; universities, out-innovated. Enter: the coding bootcamp. The idea was simple: if you already knew what you wanted — a six figure job at Amazon — you could forgo the 4-year Computer Science degree, pay just 10 or 20, or 30 thousand dollars, and “learn” to code in a fraction of the time — usually 12 weeks or less. By stripping out general education requirements and teaching to the test (tech interviews often ask the same set of questions), they promised a streamlined, more efficient, and more accessible route to the promised land. At their peak, hundreds of bootcamps graduated nearly half as many students as 4-year Computer Science programs and took in north of $200 million a year.

But as they grew, as they began appearing on billboards and in TV commercials and attracting a wider and wider group of students, they began feeling many of the same strains as universities. First, they realized their students were lending at least as much of their reputation to them as they were to students. In other words, if one of them failed an interview or was fired, that employer or recruiter would remember and associate their performance with the bootcamp on their resume, making it harder for future students to land jobs there. Silicon Valley, after all, is a small world. Most students are looking for work from one of just five companies. Thus, they began carefully guarding their reputations, selecting only those applicants who enhanced their brand. One bootcamp, Hack Reactor, for instance, has an acceptance rate of just 3% — about the same as Harvard. Another problem was a lack of teachers. Also like universities, bootcamps struggled to compete with the salaries on offer by Big Tech, often resorting to hiring their own recent graduates who couldn’t find work elsewhere, inflating their employment figures in the process. Finally, they faced pressure to raise tuition. Initially, bootcamps pitched themselves to investors much like WeWork did: as software companies. They could develop their curriculum once and then churn through students with zero marginal cost. But, like WeWork, reality turned out to be more complex. In practice, bootcamps are either little more than classrooms filled with textbooks — in which case, they don’t offer much that those same textbooks alone can’t for a fraction of the cost — or they’re highly-structured, personable, guided experiences in small, intimate settings — in which case, that’s expensive! As the number of students grows, you’ll need to keep hiring new teachers, by definition, to maintain the same student-to-teacher ratio. If anything, the marginal cost went up. Later students, attracted by their increasingly “get rich quick-syle” marketing, had less familiarity with and less interest in programming and therefore required more guidance than their earlier, more intrinsically motivated peers. Just to maintain the same outcomes, therefore, bootcamps had to raise prices. And that’s a problem when students are paying out of pocket! Bootcamps wanted in on those sweet, taxpayer-funded federal student loans. But, as unaccredited institutions, they weren’t eligible. Meanwhile, for reasons we’ve covered in previous videos, colleges found themselves short on cash. They wanted — needed — a slice of that sweet bootcamp revenue. But they aren’t very good at recruiting non-traditional students. And the Higher Education Act of 1965 prevented them from paying third-parties for bringing them students after those incentives led to aggressive and misleading marketing. …Except that in 2011, the Department of Education created a loophole: schools could pay third parties to bring them students if those payments were part of a larger “bundle of services,” of which recruiting was only one part. In other words, as long as a school didn’t call these payments “kickbacks,” they now had a green light. A university would “partner” with what they called an OPM — an “Online Program Manager” — this was the bootcamp. The OPM would create the curriculum, hire the teachers, and recruit the students, which the university would then slap its name on, usually for a 40% cut. As far as the government was concerned, these programs were accredited, so bootcamps got access to student loans. Students saw a trusted 100-year-old brand name, not some random fly-by-night bootcamp. And universities could get away with what was otherwise illegal. Put differently: bootcamps became more and more like the “unwieldy,” “inefficient” 4-year universities they originally sought to “disrupt,” until they were swallowed whole by those very same universities. Now, this more-or-less worked through the pandemic, when growth in the tech industry was off the charts and companies raced to hoard as much labor as possible. In one 2023 Wall Street Journal profile, a woman said she was paid $190,000/year by Meta to do… almost nothing — presumably because aggressive hiring at the time was rewarded by investors, and competition for talent fierce.

Then came the inevitable crash. Nearly half a million tech workers were laid off in 2023 alone. Another quarter million lost their jobs once in 2022 and again in 2024. In total, that’s roughly the number of U. S. manufacturing jobs lost from the early 2000s “China Shock. ” The unemployment rate in tech is now higher than the national average. Recent graduates have had their offers rescinded. And many professors, including this one at Berkeley — one of the most well-known Computer Science programs in the nation — report that even their best students can’t find jobs. Bootcamps are doing even worse — if they still exist at all. One, called “2U,” declared bankruptcy last year. Launch Academy announced it would “pause enrollment. ” And Dev Bootcamp, which was acquired by education giant Kaplan, closed its doors forever. It all happened remarkably fast. It seems like only yesterday that tech was “the future. ” Not so long ago, the White House predicted a shortfall of 1 million STEM workers by 2022. Instead, there are fewer software developers in the United States today than there were six years ago. How did this happen? Well, it’s not rocket science: “Learn to code,” meet another 3-word mantra: “supply and demand. ” Recall that in just ten years, the number of Computer Science grads at Berkeley increased by over a thousand percent. This was always unsustainable. Now, to be clear: “Learn to code” didn’t cause this recent downturn. In 2022, U. S. interest rates rose to their highest level in 15 years — from nearly 0% to well over 5. And the tech industry is unusually exposed to this number — making software requires massive up-front investment. When the cost of borrowing goes up, making software becomes a lot more expensive. But there’s no doubt that the over-supply of programmers — driven by “Learn to Code” — made workers more expendable in the eyes of employers. When programmers are so abundant, it’s a whole lot easier to fire them on demand — treating them as a spigot they can quickly turn on when needed and off when not, rather than a resource to invest in and retain through thick and thin. Every industry, of course, has ups and downs. And brutal as this downturn has been, growth will eventually return — the Bureau of Labor Statistics still expects computer-related occupations to grow 11% by 2033. The same, to be fair, cannot be said of all fields. Still, a starry-eyed 18-year-old back in 2016 could be forgiven for thinking that tech was somehow… different. For years, common sense had been thoroughly drowned out by the increasingly religious zeal of “Learn to code. ” “Learn to code” was more than a gentle encouragement to consider programming as a potential career. It was an all-encompassing vision of “the future,” unbounded by the laws of economics. That 2012 article from earlier, for example, compared coding to reading and writing — a surprisingly common sentiment at the time. The message wasn’t that programming was a useful skill, it was that soon, everyone would need to program. Otherwise serious people argued that everyone from electricians to teachers to insurance adjusters would be coding in the not-so-distant future. If you accept that premise, one thousand percent growth in Computer Science graduates doesn’t sound so crazy. But if programming is not a universal skill like reading or writing, if it’s still just one, admittedly often lucrative career among many, then it just means a thousand percent more competition for a limited number of jobs. Today, there are something like 1. 9 million people employed in tech. Even if we double that estimate, it would still only represent 2. 3% of the U. S. labor force. Programming simply could never have absorbed all seven million unemployed Americans. Never mind that someone still needs to manufacture the keyboard you program with, or feed the person who manufactures that keyboard, or repair that person’s internet. Tech is also at least as volatile as any other industry. The sudden and stratospheric rise of AI is a prime example.

These ups and downs become clear when we zoom back in time. This recent boom has just been so unusually long that, until recently, many young people had never experienced anything else. But “learn to code” wasn’t just at odds with Econ 101, it was also ignorant of, or at least indifferent to, the wide diversity of human interests, talents, personalities, and life circumstances. Even the most accomplished programmer will tell you: building software can be quite difficult. Like any skill, not everyone will be great at it. Nor, for that matter: enjoy it — for the same reasons not everyone enjoys working in an office, or sitting down all day, or staring at a computer, or working on a team, or solving complex problems on a deadline. This is all quite obvious when you replace “code” with “repair medical equipment,” “study epidemiology,” “become a nurse,” or any other job. Overlooking the wide spectrum of human skills, interests, and circumstances was convenient. And although it sounded empowering as a soundbite, it was often much closer to exploitative, in practice. “Learn to code” reduced people to interchangeable units of labor. Coding bootcamps represented the most distilled version of this one-size-fits-all “solution”: promising that single mothers, recently laid-off factory workers, and well-heeled highly-computer-literate 18 year-olds could all be funneled into a bootcamp and emerge, 12 weeks later, as Computer Scientists making $150,000/year. Instead, many single parents working two jobs earnestly enrolled in bootcamps only to discover they offered minimal support for and couldn’t accommodate non-traditional students. Former general contractors struggled to keep up with their peers who had prior coding experience. And since they were told learning to code is “so incredibly easy,” they blamed themselves. Thousands graduated with nothing more than a $20,000 hole in their bank account, 3 months of lost wages, and a line on their resume that, in the eyes of many employers, worked against, not for them. Others, of course, were luckier. But, perhaps most humbling of all, even many of them ultimately fell victim to “Learn to Code. ” They did exactly as they were told, set their true passions aside and pursued coding as a career, were admitted into a selective bootcamp or Computer Science program, studied hard, were fortunate enough to land a competitive, highly-paid job at Amazon, and after all that, were still one of its, say, 18,000 employees laid off with the stroke of a pen on January 5th, 2023. No matter how many times it's repeated, high schools it’s taught in, “learn to code” is no more a magic solution to economic uncertainty or job insecurity than “learn to service wind turbines” or “learn occupational therapy” — demand for which, by the way, is expected to grow by 60 and 22%, respectively — far faster than software developers. Everyone, everywhere is at the mercy of the labor market. The only solution — the best we can do — is be adaptable and invest in non-industry-specific skills like problem solving. Today’s sponsor, Brilliant, can help you do just that — spread your eggs across many baskets. Brilliant is the perfect sponsor for this video because they believe, as I do, that what’s truly valuable in this fast-moving era we live in are higher-level skills like critical and strategic thinking, creativity, and problem solving. Programming languages come and go. We forget the formulas and facts we read years ago in a textbook. But the way you approach and solve problems stays with you forever. Brilliant takes that approach and applies it to logic, puzzles, data analysis, and geometry. They have courses on “Real-World Algebra,” “Thinking in Code,” “Scientific Thinking,” “Everyday Math,” “How AI works,” “Geometric thinking,” and much more. Sure, you’ll learn to program in Python, to calculate probability, and analyze data. But you’ll also sharpen your mind, learn to think like a programmer, scientist, or mathematician, and develop a powerful learning habit along the way. Ironically, by focusing a little more on the process and a little less on the outcome, you actually get better results: by making learning fun and interactive, rather than all about “getting the right answer,” Brilliant’s

beautifully designed lessons make you want to keep coming back for more. To check out Brilliant completely free for 30 days, click the link on screen now or in the description below — that’s Brilliant. org/PolyMatter. Doing so will also get you 20% off an annual premium subscription. Go, pick what interests you most, and start learning today!