A specification is a guess wearing the costume of a plan. It's written in the confident, declarative voice of something that knows what it's doing, but underneath, every line is a prediction about how the world will behave, and predictions about the world are mostly wrong. The longer and more detailed the spec, the more wrong predictions it contains, all stated with the same false certainty. A prototype, by contrast, is a fact.
This is why I love the overnight prototype: the crudest possible version of an idea, built fast enough that it can fail honestly while you still have the energy to learn from it. It's not about cutting corners or being sloppy. It's about compressing the distance between asking a question and getting an answer. A spec answers questions by guessing. A prototype answers them by colliding with reality, which is the only authority that counts.
Most ideas die on contact with a real prototype, and that sounds like a tragedy until you realize it's the entire value. The idea that dies in a night cost you a night. The same idea, protected inside a spec and built carefully over three months, costs you three months before it dies the same death. The prototype isn't killing good ideas. It's killing bad ideas cheaply, early, before they've eaten your year.
There's a particular kind of self-deception that long planning enables, and fast prototyping defeats. While an idea exists only on paper, you can love it freely, because nothing has contradicted it yet. The plan is perfect precisely because it hasn't met anything real. Building the crude version is how you fall out of love responsibly — you see the thing fail at the thing you were sure it would do, and the infatuation breaks before it's expensive.
The objection I hear is that a rushed prototype gives you a misleading answer, because it's too crude to represent the real idea. Sometimes that's true. But more often the crudeness is exactly what makes the answer trustworthy, because it strips the idea down to its load-bearing claim and tests only that. If the core doesn't work even in the simplest version, no amount of polish on a full version was going to save it. You learned the only thing worth knowing.
What the overnight prototype really buys you is the right to fail cheaply, over and over, until something doesn't. Teams that prototype relentlessly aren't reckless; they're the opposite. They refuse to commit real resources to an untested belief, so they test it for the cost of a night before they spend a quarter. From outside, the constant building can look like thrashing. From inside, it's a disciplined refusal to be fooled by a plan.
This also reshapes how you think about being wrong. If being wrong takes three months to discover, you'll avoid it, defend against it, and rationalize away the early signs, because the cost of admitting it is enormous. If being wrong takes a night to discover, you can be wrong ten times a week and call it progress, because each wrong answer was cheap and each one narrowed the search. Cheap failure changes your whole relationship to the truth.
So when you have an idea and the urge is to write it down properly, plan it out, get it right — consider building the ugliest working version of it tonight instead. Not the real thing. The crudest thing that can prove or kill the core claim. You'll learn more by morning than the spec would have taught you in a month, and whichever way it goes, you'll know something true instead of believing something tidy.