There's a piece of conventional wisdom in startup culture that goes something like this: ideas don't matter, execution is everything. This is not only wrong, it's harmfully wrong. Ideas are more important than execution, and I think the people repeating this mantra have been pattern-matching on the wrong thing.
Let me start with a basic observation. The technical side of anything (science, engineering, business) is itself made of ideas. When someone invented the wheel, the hard part was not carving the thing. It was realizing that a circular object rotating on an axle could convert a dragging problem into a rolling one. Once you have that insight, any competent woodworker could execute it. The gap between "no wheel" and "wheel" is an idea gap, not an execution gap.
Patents are ideas. Trade secrets are ideas. The entire legal infrastructure we've built to protect competitive advantage is, at bottom, a system for protecting ideas. This should tell you something about where the value lies.
There's a related point the ancient Chinese philosophers understood well in the debates about míng and shí, names and reality. Naming things is not mere labeling. A name confers a higher-dimensional abstraction onto a thing, and that abstraction opens doors. When people called a certain architecture of backpropagation a "neural network" and the broader field "artificial intelligence," that wasn't idle metaphor. Those names imported a frame: the frame of cognition, of biology, of something that learns. And that frame attracted talent, funding, and ambition that "nonlinear function approximation" never would have. The name was an idea, and it mattered enormously.
Or consider AlexNet. The key insight behind AlexNet was that GPUs, hardware designed to render video game graphics, could be repurposed to train convolutional neural networks at a scale nobody had tried. That's an idea. And it triggered the entire deep learning revolution.
But look at what happened next. Krizhevsky, Sutskever, and Hinton had to make a series of engineering choices: ReLU activations, dropout regularization, aggressive data augmentation. These are usually filed under "execution." But each one was itself an idea. ReLU was the insight that a dead-simple activation function would train faster than the theoretically elegant alternatives. Dropout was the insight that randomly disabling neurons during training would prevent overfitting better than any explicit regularizer. These weren't mechanical steps that followed inevitably from the initial idea. They were new ideas at a finer granularity, each one a judgment call that could have gone wrong. The "execution" of AlexNet was a cascade of ideas all the way down. Without the directional idea at the top, there is nothing to optimize. But without good ideas at every level below it, the optimization fails too.
Aristotle distinguished between form (μορφή) and matter (ὕλη): matter is the raw material, form is what gives it structure and purpose. A lump of bronze is matter; the shape of a statue is form. And here is the crucial point: matter, for Aristotle, is not the important thing. It is the opposite. Matter is mere potentiality (δύναμις), the thing that could become something but hasn't yet. Form is actuality (ἐνέργεια), the thing that makes a thing what it is. The bronze is nothing until it receives the shape. The actual is always prior to the potential.
But most people miss a further subtlety: matter is not a fixed layer at the bottom. Matter itself decomposes into matter and form at a lower level. Bricks are the matter of a house, but a brick is itself clay shaped into a cuboid, clay is mud given a certain consistency, and so on. At every level of decomposition, it is form that does the explanatory work: form is what makes a thing what it is rather than a heap of raw ingredients.
Execution is the matter of a startup, in precisely this Aristotelian sense: the raw potential that has no direction until an idea gives it one. Ideas are the form. And if you look closely at execution itself, it dissolves into a series of smaller decisions: what to prioritize this week, how to structure the database, which customer segment to target first. Each of which, done well, requires a specific insight about what matters in that moment. That insight is an idea. Go one level deeper and you find the same thing: each micro-decision decomposes into still finer judgments, each requiring its own insight. Execution is not the opposite of ideation. It is ideation at a finer granularity. The people who are good at "execution" are, in fact, people who are good at generating ideas at every level of detail.
Follow this all the way down and you reach a strange conclusion: execution, as a separate thing, doesn't really exist. It is just the recursive unfolding of ideas. The woodworker still has to carve the wheel. But every cut he makes is guided by a judgment about where to cut next, and that judgment is an idea. Execution without good ideas is labor. Execution with good ideas is leverage.
This has a practical consequence. Good ideas and bad ideas differ enormously even at the conception stage, and the difference is not vague or hand-wavy. It's fine-grained. A good idea tends to already be more practical and more complete in the mind of its conceiver. From the very beginning, a good startup idea may account for where it sits in the competitive landscape, what its natural moat is, why the timing is right, and how the business model works. A bad idea, by contrast, tends to be an abstraction floating free of such details, a vision statement in search of a plan.
And if execution is really ideas all the way down, then this gap only widens over time. Someone who can generate good ideas doesn't stop having them after the first one. During execution, they keep producing more ideas about the product, the market, the technical architecture, the hiring strategy. Each of these micro-ideas compounds. The project ends up operating far better than one run by someone with an average initial idea, because it's being fueled by better judgment at every turn.
So why do people keep saying ideas don't matter?
I think it's because they're fixating on the wrong population. They're looking at people who have grandiose ideas but never build anything, the all-talk, no-code crowd. And it's true: among people who can't build, the ones who merely fantasize are indistinguishable from (and arguably worse than) the ones who at least go try something. They're worse because they don't even get the benefit of learning by doing.
But this observation has no relevance whatsoever to people who actually build things. For a doer, someone who is going to execute regardless, the difference between starting with a good idea and starting with a bad one is enormous. The advice "don't worry about ideas, just execute" is like telling a navigator "don't worry about the direction, just row harder." Rowing matters. But direction matters more.
This is becoming truer by the day. And this is the part I want to be most precise about.
In the age of AI, execution is getting dramatically cheaper. The cost of building a prototype, writing code, generating content, testing a market: all of this is collapsing. What used to take a team of engineers six months can now be prototyped by one person in a weekend. This changes the relative scarcity of ideas and execution in a fundamental way.
For most of startup history, the conventional wisdom had a point. Ideas were relatively abundant (anyone could have one at a cocktail party) and execution was genuinely scarce. Building software was hard, expensive, slow. The bottleneck was execution, so execution commanded a premium. "Ideas are cheap, execution is everything" was a reasonable heuristic for a world where shipping was the hard part.
That world is ending. When AI can generate code, design interfaces, draft marketing copy, and automate testing, execution stops being the bottleneck. The scarce resource shifts to knowing what to build: which problem to solve, which market to enter, which architecture to choose, which trade-offs to make. These are all ideas. And scarce resources are what determine value.
The people who will build the most important companies in the next decade won't be the hardest workers, though they'll work hard. They'll be the ones who see what to build, who look at the world and notice the right sticking door. [1]
Notes
[1] This phrase is borrowed from Paul Graham, who uses the analogy of a carpenter noticing a sticking door to describe how technically skilled people perceive missing or broken things in the world. The point is the same: seeing what to build is itself the skill.