What if we’ve designed education in the wrong order?
James Dyson built 5,126 vacuum prototypes that didn’t work.
Five thousand, one hundred and twenty-six.
For fifteen years, he built prototype after prototype. Some failed completely. Others got him a little closer. Then came prototype 5,127.
That one worked.
Now imagine Dyson had known from the beginning that number 5,127 was going to work.
How differently would he have treated the first 5,126?
Would prototype 37 have felt like failure? Would he have hidden prototype 2,000? Or would every failed attempt have meant one thing:
I’m getting closer.
Now imagine we told an eighteen-year-old something similar.
Imagine that becoming truly capable required exactly 500 mistakes. Five hundred bad ideas, failed experiments, broken prototypes, arguments you lose, projects that embarrass you, and things you build that simply do not work.
But there’s a catch: you know, with absolute certainty, that attempt 501 is going to work.
How differently would you treat the first 500?
Would mistake 37 feel like failure, or would it feel like progress? Would you hide mistake 214, or would you think, Good. Only 286 to go.
Would you spend four years trying to avoid being wrong?
Or would you start counting?
Of course, there is no magic number. But the thought experiment reveals something important: those first 500 attempts aren’t failures on the way to becoming capable.
They are how capability gets built.
We tend to assume that people become capable and then take on difficult things.
Maybe we have it backwards.
They don’t become capable and then do difficult things. They become capable by doing difficult things.
Over the years, I’ve had the chance to look at learning from several very different vantage points: behavioral science, startup communities, and now higher education. Across them, I kept noticing the same disconnect.
The environments that seemed to produce the most capable people rarely followed the sequence we traditionally associate with learning. People weren’t becoming capable because someone had perfectly explained what to do. They were trying things before they felt ready, getting stuck, seeking knowledge because they suddenly needed it, relying on other people, getting things wrong, and trying again.
Meanwhile, much of formal education still follows a familiar sequence:
Learn → Test → Do
First, we teach you what you need to know. Then we test whether you’ve learned it. Eventually, once you’ve demonstrated enough competence, you’re allowed to apply that knowledge to something real.
It makes intuitive sense. Knowledge comes first; application comes later.
But the world we’re preparing people for rarely works that way.
There is no answer key for starting a company, no rubric for navigating a crisis you’ve never encountered, and no syllabus for what to do when your strategy stops working or a new technology changes the assumptions your industry was built on. The problems that matter most are often precisely the problems for which nobody knows the answer yet.
So I’ve been wondering: What if we’ve designed education in the wrong order?
What if, instead of preparing people for reality before allowing them to encounter it, we started with reality?
The alternative I’ve been thinking about is surprisingly simple.
Do → Struggle → Learn → Repeat
Start with something real. Encounter the limits of what you currently know. Pull in knowledge, teachers, peers, and tools when you need them. Then take what you’ve learned back into the problem and try again.
The shift sounds small, but it changes the logic of education. Instead of learning so that someday you can do, you do so that you have a reason to learn.
Imagine taking an entrepreneurship course.
In the traditional model, you might study market sizing, positioning, customer discovery, unit economics, and fundraising. You analyze companies, discuss cases, and eventually develop a business plan.
Now reverse it.
On day one, your assignment is to find one person willing to pay you for something.
Suddenly, you have questions. Who should I sell to? What problem am I actually solving? How much should I charge? Why isn’t anyone responding? Is my product wrong, or am I talking to the wrong customer?
Now market segmentation isn’t an abstract concept. You need it.
There’s a well-established idea in cognitive psychology behind this. In their classic research on the generation effect, Norman Slamecka and Peter Graf found that people remembered information better when they generated it themselves rather than simply reading it.
The educational implication is simple: don’t always give people the answer before they’ve had a reason to need it.
You can see a version of this at Team Academy in Finland. Rather than separating the study of entrepreneurship from the practice of entrepreneurship, students work in teams on real commercial projects. They find customers, develop services, sell, make decisions, and then reflect on what happened with coaches.
The project isn’t simply what happens after the learning.
The project creates the learning.
Once you start doing real things, something predictable happens: you get stuck.
That’s usually the moment we rush in to help. But perhaps we shouldn’t rush quite so quickly.
Robert and Elizabeth Bjork’s research on desirable difficulties points to something counterintuitive: some conditions that make learning harder in the short term can improve retention and transfer over time. Related work by Nate Kornell and Robert Bjork found that interleaving different kinds of problems could make practice feel harder while improving later discrimination between concepts.
The important word is desirable. Difficulty for its own sake isn’t education. The goal isn’t to make learners suffer. It’s to create enough friction that they have to construct understanding rather than simply follow instructions.
École 42, the software engineering school founded in Paris, pushes this philosophy unusually far. Instead of relying on traditional lectures and professors, students encounter programming problems they don’t yet know how to solve. They try, their code fails, they search, they learn from peers, they debug, and they try again.
The interesting outcome isn’t simply that they learn to code. Repeatedly confronting unfamiliar problems forces them to practice something conventional education often tries to minimize: not knowing what to do next.
Over time, the learner becomes comfortable saying:
I don’t know how to solve this yet, but I know how to begin.
That may be one of the most valuable capabilities education can produce.
This is where I want to be careful.
I’m not arguing that we should get rid of teachers, lectures, books, or theory. Quite the opposite.
I’m arguing that we may have put them in the wrong place in the loop.
Imagine struggling with a problem for three days and then encountering a professor who can explain exactly why your approach isn’t working. You’re listening differently now. The theory has somewhere to attach. The professor isn’t answering a question someone decided you ought to have; they’re helping you answer a question you genuinely want solved.
Learning becomes pull rather than push.
And it becomes social.
Someone sees what you missed. Someone has knowledge you don’t. Someone challenges an assumption you’ve become attached to. Eventually, you have to become comfortable saying things like I don’t know, I was wrong, Can you help me?, and Your idea is better.
Amy Edmondson’s research on psychological safety offers an important insight here. In her study of hospital teams, teams that appeared to report more errors were not necessarily worse teams. One explanation was that psychologically safer environments made it easier for people to report and discuss mistakes rather than hide them.
That distinction matters.
A culture that punishes being wrong doesn’t necessarily eliminate mistakes. It can simply make mistakes less visible.
And a mistake that stays hidden cannot become shared knowledge.
So perhaps another principle of this model is:
Make being wrong cheap.
Not consequence-free. Not meaningless. Cheap enough that people surface errors quickly, examine them, and learn from them.
The faster a group can say that didn’t work, why?, the faster it can improve.
Then take what you learned and go back into reality.
Build another version. Make another decision. Run another experiment. Try another approach. Discover the next thing you don’t understand.
Do → Struggle → Learn → Repeat.
Each loop leaves something behind. Knowledge, certainly, but also judgment. You learn which assumptions to test first, when to persist, when to change direction, how to ask for help, how to recover from a bad decision, and how to begin before you feel completely ready.
Eventually, the learner doesn’t simply know more.
They need less certainty in order to act.
To me, that’s agency.
Imagine arriving at college and receiving a problem before receiving a syllabus.
Your team has twelve weeks to solve something real for a real organization.
You don’t learn accounting simply because accounting is required. You encounter a cash-flow problem and suddenly need accounting.
You don’t study negotiation because it’s Tuesday at 10 a.m. A partner says no, and now you need to understand negotiation.
You don’t study leadership because there is an exam on Friday. Three people on your team disagree about what to do next, the deadline is approaching, and somebody has to get the group moving again.
Faculty don’t disappear in this model. In some ways, they become more important.
But their role changes.
They diagnose. Challenge. Coach. Introduce theory when it becomes useful. Help students distinguish productive struggle from pointless struggle. Ask the question the team is avoiding.
And increasingly, AI could play another role in this system: giving every learner access to instruction at the precise moment they encounter a knowledge gap.
The learner attempts something first. Gets stuck. Seeks help. Learns. Immediately applies what they’ve learned. Reality provides feedback.
Then the loop begins again.
This isn’t education without instruction.
It’s instruction in response to experience.
Which brings me back to our eighteen-year-old.
Imagine that instead of spending four years trying to prove how rarely she gets things wrong, we gave her four years to become extraordinarily good at learning from being wrong.
Give her real problems. Give her teammates who depend on her. Give her exceptional teachers when she gets stuck. Give her access to knowledge when she needs it. Give her enough support that failure is survivable, but not so much that struggle disappears.
Then let her build.
When something doesn’t work, ask: What happened? What did you believe that turned out not to be true? What did someone else see that you didn’t? What do you know now that you didn’t know before? What will you try next?
Then send her back out.
Mistake 37. Learn.
Mistake 214. Learn.
Mistake 499. Learn.
There is no magical attempt 501, of course. But by then, something else has changed.
The person making attempt 501 is no longer the person who made attempt one.
She knows how to walk into ambiguity without waiting for instructions. She knows how to discover that she’s wrong without losing confidence. She knows how to find people who know things she doesn’t. She knows how to learn from reality.
She knows how to begin without an answer key.
Maybe that’s what education should be designed to produce.
Not people who make fewer mistakes, but people who get better because of them.
The purpose of education isn’t to prepare you for attempt 501.
Attempts 1 through 500 are the education.