Over the weekend, more than one person told me that reading is hard and boring.
Not kids. Adults. Capable people who run families and companies. They weren’t bragging, and they weren’t ashamed. They said it the way you’d mention you don’t care for cilantro, a small fact about themselves, nothing to argue about.
It stuck with me, because every one of them also wanted to know things. They had questions about money, health, the news, and the tools they use all day. They wanted the answers. They just didn’t want the part where you sit with a hard text and build the answer yourself, slowly, out of pieces you had to go get.
This is most people now. They want knowledge without the effort, and for the first time, there’s a machine that promises to deliver it.
What the words actually mean
“First principles” has become a LinkedIn flex. Before it got flattened into that, it meant something precise.
The Greek word is arche. Beginning, origin, source. The Latin is principium, the same idea. A first principle is a basic proposition that can’t be deduced from anything more basic. You hit it, and you stop, because there’s nothing underneath to stand on. In math, we call these axioms. Everything else is built on top.
The pre-Socratics went looking for a single one. Thales said everything was water. Anaximenes said air. They were wrong, but what they were searching for was right. Find the thing at the bottom, then explain the rest in terms of it.
Aristotle made it rigorous. Real knowledge, he argued, has to rest on first principles that themselves can’t be proven. Try to prove everything instead, and you hit one of two dead ends. You reason in a circle, where each claim leans on another that leans right back on the first, or you chase the justification backward forever and never reach the bottom. Neither one gets you to something you actually know, so you have to grant a few things at the start without proof. Those are the archai.
Euclid then showed what you could do with a floor. From five postulates, five common notions, and their combinations, he derived hundreds of geometric proofs. Every theorem traceable back to a handful of things he asked you to grant at the start. Two thousand years later, it still holds up. That’s the power of building from the bottom up.
Descartes took it further than anyone. He threw out everything he could possibly doubt, his senses, the world, his own body, until he hit the one thing he couldn’t doubt: that he was doubting. I think, therefore I am. One indubitable principle, and he tried to rebuild all of knowledge on it.
You don’t have to like where any of them landed to see the shared idea. Strip away what you inherited. Keep stripping until you reach something that can’t be reduced. Build back up from there.
The honest version, and the buzzword
Feynman is the one who kept it honest. His version of the first principle wasn’t an axiom; it was a warning: “You must not fool yourself, and you are the easiest person to fool.” He told students to work things out ab initio, from the start, from their own experience, instead of trusting the answer handed down. “Science is the belief in the ignorance of experts.” Not contempt for experts. Distrust of taking their conclusions on faith without being able to rederive them.
Then Silicon Valley got hold of it. Around 2013, Elon Musk started saying that physics teaches you to reason from first principles rather than by analogy. The example everyone repeats: the raw materials in a rocket are about two percent of what a rocket costs, so the price is a choice, not a law of nature. Build the rest yourself, and the number falls. SpaceX proved that in the most fun way I can think of. They pushed and iterated until they succeeded. And all along the way, they laughed at themselves when they failed. And rockets fail in VERY spectacular ways.
There’s something real in that. There’s also a reason “first principles” became, in one critic’s words, a vanity buzzword and a crutch for analytic incompetence. It works in bounded domains, physics, manufacturing, cost structures, and places where actual fundamentals can be found. If you stretch it past those, it curdles. Adam Tooze pointed out that Tesla’s start-from-scratch instinct can produce cars that feel “engineered by Martians,” as though a century of accumulated knowledge about chassis and brakes simply didn’t exist. Sometimes the inherited thing is right.
So the method cuts both ways. Refuse to reason from fundamentals, and you’re just copying. Refuse to respect what’s already been learned, and you reinvent the flat tire. The skill isn’t picking a side. It’s knowing which problem applies.
Reasoning by analogy, industrialized
One distinction in all this is the one that counts if you build software in 2026.
Reasoning from first principles means going to the bottom and building up. Reasoning by analogy means copying what already exists with slight variations. Most work is analogy, and that’s fine. You don’t reinvent the wheel every morning. You see a problem that looks like one you’ve seen, and you apply the shape of the old solution. It’s fast, it’s cheap, and it’s right often enough to run a career on.
Now look at what an LLM does.
It’s the most powerful analogy engine ever built. It has read more code, more text, more solved problems than any human could in a thousand lifetimes, and it answers by finding the nearest pattern and reshaping it to fit. That’s not a knock. It’s why these tools are so good at code, they’ve seen it all before. Underneath the magic of its prediction is the most sophisticated “what usually comes next” machine we’ve made.
Which means “just prompt it” is reasoning by analogy at an industrial scale. You ask, and the machine returns the average of everything anyone has already done that looks like your question. For the enormous category of problems that are basically solved, this is a gift. For the problems that aren’t, it’s a trap, because the machine will still answer confidently, with the closest analogy it has. And the closest analogy to a genuinely new problem is, by definition, wrong.
The model can’t tell you when you’ve left the territory it has a map for. Only you can. And you can only do it if you can reason from first principles yourself, which is the one thing the tool can’t do for you.
Descending the stack
We build software in layers, so we never have to think about the bottom. The framework sits on the runtime, the runtime on the kernel, the kernel on silicon, and most days you live near the top, and it’s wonderful. Abstractions are how we manage complexity without having to hold the whole machine in our heads.
AI is now the top layer. The newest abstraction, sitting above all the others. Describe what you want, and it assembles the lower layers for you. The same trade we’ve always made, one more level up.
The trade is good until the day it breaks. The query is slow, and the ORM won’t tell you why. The container behaves differently in production, and the cause is three layers down. The model wrote something plausible that fails in a way it has never seen, so it can’t pattern-match its way out. On that day, the only people who recover are the ones who can climb down the stack and reason about what’s actually happening at the bottom.
That ability doesn’t come from prompting. It comes from having gone to the bottom before. From reading the slow, hard texts, the source, the spec, the paper, the thing that explains why the layer exists, instead of just how to call it. The effort people told me they’d rather skip is the exact effort that builds the floor you stand on when the abstraction gives out.
The part nobody wants to hear
You can rent the answer. You can’t rent the understanding.
The people who told me reading was hard and boring were right. It is. Building knowledge from first principles is slow and uncomfortable, and there’s no version where it isn’t. That was true for Aristotle, and it’s true now. What changed is that we finally built a machine good enough to make the shortcut look like the real thing.
It isn’t.
The machine gives you the analogy. The floor is still yours to build, one hard text at a time, and it’s worth more than it’s ever been, precisely because almost nobody wants to do it anymore.
Reading is hard and boring. That’s not a reason to stop. It’s the reason it works.
What’s the last thing you read that was genuinely hard, the kind you had to fight through? I’d like to know if I’m the only one who still does it on purpose.