A few months ago, a friend of mine got tired of his job and decided to learn to code. Suddenly, he’s a professional programmer. There are several reasons this happened — he’s a smart and curious guy, and he really wanted a better career. And there’s one specific reason it happened when it did: Lambda School.
Lambda School was not the first coding bootcamp, but it’s the first one to recognize that one of the big problems in education is mismatched incentives. State and federal governments direct a torrent of subsidies to student loans, and they’re largely indifferent to the quality of the collateral. This gives schools an incentive to stretch the definition of a degree as far as possible. If Fannie and Freddie gave anything called a “house” a hundred-thousand dollar mortgage, you’d expect a lot of shacks.
And that’s just what we get, at least in the legacy education business. We vastly overtrain people for dying careers like journalism, or non-careers like English. This might be virtuous, but it’s confusing: if “school” can be an investment in a high-paying career but also a four- (or five-, or six-) year party with some Wittgenstein and Proust thrown in, “school” ceases to be a useful term.
That merely charging differently can become the core of a business reveals just how broken conventional higher education really is. Unlike a normal school, or even a typical coding program, Lambda School charges a percentage of post-graduation salary — 17% of total comp for two years, but only for students who earn at least $50k, with total payments capped at $30k.
This is perhaps the sanest deal in education, except for natural autodidacts. It’s a very expensive bootcamp, if and only if it turns out to have been worth it.
Americans are chronically overschooled. Almost 70% of high school students enroll in college, and about 55% will actually manage to graduate within six years, but that’s largely a signaling game. When everyone has a degree, a degree becomes a minimum requirement for even low-end employment.
But a lot of us have other things going on in our late teens and early twenties, and the opportunity cost of school for someone with ambition is enormous. Bryan Caplan’s education book has plenty of evidence for this, both statistical and anecdotal (perhaps the best single thought experiment here: why are students relieved when their teacher is out sick? They’re getting ripped off! Unless what they’re paying for is the credential, not the knowledge.)
If we’re overschooled, why do even more? Because there are still some underserved markets. A while ago, a study showed that high-achieving students from poor families are especially unlikely to apply to colleges. One place where that might be especially common: a small town in a poorer-than-average state, where a large number of students have social obligations that wipe out a few years during prime college-going time. A town like, say, Springville, Utah, where Lambda’s founder, Austen Allred, is from.
So even if the country in the aggregate is over-credentialed, that doesn’t mean that selling education is a bad business. In fact, unbundling education from credentials, and selling the low-cost/high-value component for what it’s worth can be a great business.
The lifecycle of any bundled product goes like this:
- The bundle comes into existence when sellers realize that there’s a market for product A and B (and C and so on), but that selling A+B together for a slight premium is a slightly better deal for most people who want either.
- The bundle grows continuously, in both volume and price. As demand tops out, the big growth is in pushing price rather than increasing customer count.
- Eventually someone realizes that a bundle of A through Z is pretty expensive compared to just selling Q, so they disaggregate the bundle, focus on Q, and make… a bundle.
This is a cyclical phenomenon because it’s a lot easier to add to the package than to remove from it, so you need a fierce level of minimalism to compete. It’s happened in retail, transportation, retail stock brokerages, and online content. It makes sense that the unbundlers are typically challengers, not incumbents: if the two good options are a) raise quality 1% and raise prices 5%, or b) reduce quality by 80% but cut prices by 90%, it’s hard to choose b) when a) has been the consistently smarter option.
There are two main lines of criticism for Lambda School. Either it’s not enough or it’s entirely too much.
On the not-enough side, Lamda omits both some of the hard theory that a CS major needs and some of the softer stuff they’re increasingly expected to absorb.
I can see a good argument for teaching more CS theory, but not to everyone. It’s helpful to know Big-O notation and the theoretical limits of what a computer can do — but not if you’re just building an internal portal that helps employees request vacation days and access their W-2s. (Edit 3/5/2019: I have been informed that Lambda School now does include this kind of theory in their curriculum.) Big tech companies are notorious for asking complex theoretical questions at interviews, but that strikes me as more of an intelligence and interest filter than a real qualification. It’s like the theory of jocks in finance: you don’t hire an athlete to do banking because they can run really fast; you hire them because the traits that lead someone to train for a sport are good traits to have in a demanding job.
It could be that every programmer eventually runs into hard CS problems, and theory does come in handy from time to time. (I still remember how proud I was when I converted a trial-and-error approach into a linear algebra problem and realized that it was unsolvable, rather than merely tedious.) But there’s a large space of potential programming tasks that just don’t need that knowledge, or that need it so rarely that it’s not worth developing “just in case”. An Electrical Engineering PhD is nice to have, but a journeyman electrician can do valuable work with a more limited and pragmatic understanding of what electrons do.
The soft-skills problem is thornier. A common criticism of big tech companies is that, if they were run by liberal arts majors, they’d behave differently. Consider this scenario: you work for a sporting-goods company, and they ask you to build a website that allows customers to order guns and ammunition. You need to set up some form to check their ID and run a background check before delivering the gun to a local gun dealer.
If you’d taken an ethics class, you’d know this is unacceptable: you’re helping the state interfere with law-abiding citizens’ rights to their own protection! At least, that’s what you might learn if your “Ethics in Computer Science” class had been taught at Hillsdale. If you went to Oberlin, maybe you’d think something different.
This is not some wishy-washy po-mo claim that there’s no such thing as objective morality. (Obviously I think my opinions are correct.) The problem with ethics-in-CS is that all the topics being discussed are live political issues, where there generally isn’t agreement on the right thing to do. I think most of us would agree that it’s ethical to try to limit carbon emissions, and that it’s ethical to help the poor. But, given that most of global emissions growth is a side effect of poor countries catching up to rich ones, thinking deeply about ethics doesn’t actually answer any questions, it just generates harder questions. If you could flip a switch that sent China’s standards of living back to where they were in 1980, you’d a) solve a bigger chunk of global warming than the Green New Deal, and b) cause more human suffering than anyone in human history. Oops!
A good ethics professor can convince you not to work on software for combat drones, but a thoughtful one might argue that if drones had somehow been invented before conventional warfare, the first person to propose actually sending a bunch of people to get shot at would have been dismissed as a dangerous crank.
Ethics is hard. It sounds fun and I can see why someone might enjoy teaching the class or taking it. But if you’re trying to train people well, I can also see why you’d skip it.
Team too-much has a different argument: Lambda School is hard to complete. I don’t have solid data on this, just anecdotes from around the Internet: the big complaints about Lambda School are about the lack of support, either for students who are struggling or for students who run into personal problems that prevent them from finishing school.
Part of the higher education bundle is a sort of welfare system for students who run into personal problems — American schools are much more likely than workplaces to work around a student who can’t complete the workload because of a personal crisis. Given that most of these complaints are one side of the story, I’m not sure what Lambda School’s policy is. If they’re smart, their rule should probably be that a student with a serious medical issue, family problem, etc., can defer completion while they deal with it. This would make Lambda School less full-service than a school that provides students with food and lodging, but that’s part of why students aren’t paying the same tuition they would at a competing institution. After all, bureaucratic bloat is a huge contributor to college’s runaway costs. Those “chancellors of student experience” don’t help students face their personal problems for free (and it’s not obvious the expensive regular system is that effective, either).
When people say that Lambda School expects students to sink or swim, or that instructors ask students to Google error messages before asking about them, they’re giving the school a positive review. 80% of programming consists of learning how to use Google and Stack Overflow effectively. This takes a certain mindset, but it’s learnable. When a beginner looks at broken code, the natural thing is to think of it as something that’s broken in the way a coffee mug shatters if you drop it on the floor. But, to a good developer, code is actually broken in the way that a puzzle is — incomplete, but you have the pieces necessary to fix it.
Extending the Model
Education is a market so broken that simply charging differently can serve as the core of a new business. If you look at other parts of the economy that seem similarly broken, you might reach the same conclusion.
Take healthcare. Our healthcare system is built around a loophole in World War Two-era wage controls, but if you’ve checked the news recently you may be aware that Hitler and Hirohito were actually defeated some time ago. Could we charge better? Could we replicate the Lambda School incentives-based model?
We sure could. Instead of selling access to healthcare, Lambda Health could offer life insurance policies and then subsidize the preventative care that prevents those policies from being used. You could set it up as a policy that pays 10x your annual income if you die between 30 and 50, then slowly drops from there. Then, your healthcare provider isn’t in the(backwards) business of selling you pills or surgery; they’re in the business of providing you with whatever treatment keeps you alive the longest.
(This is a basic sketch, and there are all sorts of incentive problems. In a cyberpunk dystopia scenario, they might realize that the optimal business strategy is to reduce accident risk and caloric expenditure by giving everyone high doses of opiates or something. So you’d really want to have a system where there are payouts from the insurer based on your achievement of various milestones. Still deadlifting a lot at 50? They cut you a check. Decide to take up smoking? They don’t.)
Real estate might be amenable to the same sort of cleverness. I remain obsessed with the idea that, in the US, we’ve chosen to bundle together purchases of shelter and highly-leveraged exotic bets on interest rates and local labor markets. This does not seem like the system you’d derive if you started from first principles, though it’s a hard system to iterate out of. Lambda Home might rent you an apartment paired with unemployment insurance, and charge you a higher unemployment insurance premium when wages where you live go up.
The Lambda School team is currently a little bit busy fixing the education system, so they probably don’t have time to solve healthcare and housing, too. Maybe after the IPO.
In the meantime, what will they do next?
They’ve already said they plan to expand into nursing schools, which makes plenty of sense. Medicine, like programming, is a field that requires just a few specialists but many hard-working generalists.
But I’m a finance guy, so I’m naturally less interested in product than I am in finance, and the financial perspective on Lambda School is really, really interesting. Because they don’t usually charge upfront, they aren’t quite in the business of selling a product with some cost of goods sold; they’re actually making an investment and acquiring a financial asset that matures over time. Specifically, Lambda owns preferred stock in its alumni: they get some of the upside, but it’s capped.
From one perspective, it’s very strange to look at a tech startup whose balance sheet is just a bunch of esoteric financial securities. On the other hand, they’re recreating an important part of the education system. Yale has billions of dollars invested in private equity, venture capital, real estate, and timber, and they make a lot of money from alumni donations which correlate with alumni income. As it turns out, it would be weird for an education company not to own a diversified portfolio of alternative assets.
What Lambda School should do next is look backwards a little: they were funded by Y Combinator, whose key economic value-add is that they identify promising outsiders and spend three months turning them into insiders. (They do other things, too, but in terms of dollar value their most important service is that by the end of the program, VCs are approaching the founders, rather than the other way around.)
So, Lambda’s next move should be a seed accelerator. That’s right: having been funded by Y Combinator, they should turn around and compete with them. As a concession to politeness and sanity, they can confine themselves to funding only startups that are founded outside of San Francisco. YC makes companies move to the Bay, and some people don’t want to do that — in fact, the people who don’t want to do that are also the sorts of alternative students who are more likely to go to Lambda School than to get a four-year degree. While the mean valuation of Bay Area-founded startups will likely be higher than startups in the rest of the country for a while longer, moving to Silicon Valley isn’t the obvious call it was a couple years ago. Given limited housing stock and generous comp from Big Tech, a large share of the wealth creation in San Francisco goes straight to landlords.
Lambda School has lots of data on students. They know who completes assignments on time, who works well in teams, who will pull an all-nighter rather than miss a deadline, who can handle a crisis without melting down. Lambda School is a four-month program, and with Lambda Fund it would also be a four-month VC pitch.
Educators like to believe that they change their students for life, and for the better. They seem to stop believing this as soon as it’s time to negotiate wages — at that point, they’re focused on job security (for themselves) and pay-for-seniority. The Lambda School model is a chance for the American educational system to have skin in the game. Let’s see if they take it.
 There are lots of self-identified autodidacts, but the completion rate of free Coursera courses shows that most can’t walk the walk. Identifying as an autodidact online is about as rigorous a qualification as saying you’re Irish on St. Patrick’s Day. (And I say this as someone who tapped out of Andrew Ng’s class halfway through.)
 The NYT article mentions a few locations: “Bridgeport; Memphis; Sacramento; Toledo, Ohio; and Tulsa, Okla.” Not really Mormon country. It’s quite likely that the admissions staff at BYU is unusually understanding of twenty-year-old applicants with a two-year resume gap after high school, but that still limits students’ options a bit.
 This fits a common pattern in successful startups: they solve an especially pressing version of a general problem, then generalize from there. Is it a coincidence that a successful college alternative was founded by a Mormon, that the best search engine was invented by grad students who had to find lots of CS papers, that Salesforce’s sales-enhancing software was conceived by a busy salesman, that LinkedIn’s founder has a network so broad he can’t keep track of where everyone’s working? My networking problems are probably about one or two milliHoffmans, but the product that works great for him works fine for me.
 A modest addition to the bundle nets some incremental subscribers and justifies a price hike, and since existing subscribers are less price-sensitive than new ones, the price and scope of the bundle tends to expand.