The Most Valuable Thing in AI Is the Data Nobody Has Yet

Start with the part that should reorganize how you think about all of this.

Everybody has the same models now. The frontier AI you can buy is, give or take, the frontier AI your competitor can buy and the frontier AI a stranger in another country can buy. The model is not the scarce thing anymore. It is becoming a utility, like electricity or bandwidth, and you do not build a defensible business on having the same electricity as everyone else.

So if the model is not the edge, what is. The answer is almost embarrassingly simple once you see it. The edge is information the model has never seen. Data that does not exist anywhere yet, because it has not been collected, because nobody went out and made it. In an era where deriving answers from existing data has become nearly free, the rare and valuable act is producing new data in the first place. The whole game has quietly moved from who has the best model to who is feeding it something no one else can.

I wrote a piece before this one about why everything an AI produces is derivative, derived entirely from the past, a function built on top of human-made data with no independent source of newness of its own. This is the other half of that thought, and the more hopeful half. If the machine can only ever take a derivative of an existing source, then the source is everything. So let me go deeper into where new sources actually come from, why creating one is different from reiterating an old one, and what we are actually doing about it at Entoura.

Combining, exploring, and the one the machine cannot do

The cognitive scientist Margaret Boden spent a career thinking clearly about what creativity even is, and her framework is the most useful one I know for cutting through the AI hype. In her book The Creative Mind: Myths and Mechanisms, she splits creativity into three kinds, and the split maps almost perfectly onto what these tools can and cannot do.

The first kind is combinational. You take familiar ideas and put them together in an unfamiliar way. A metaphor, a mash-up, a fresh pairing of two things that were always lying around. The second kind is exploratory. You take an established space with its own rules, a style, a genre, a method, and you push into corners of it that had not been filled in yet. You find a new move inside an existing game. The third kind is transformational. You change the rules themselves. You break the space open and make a new one that could not have existed under the old constraints. This is the rarest kind, and it is the kind that produces actual revolutions rather than variations.

Here is the thing. Generative AI is genuinely, dazzlingly good at the first two. Combinational creativity is practically its native language, since recombining the patterns in its training data is exactly what it does. Exploratory creativity it can do too, filling in unexplored corners of a known style with real fluency. If your work lives in those two kinds, the machine is an incredible partner and you should be using it hard.

But transformational creativity, changing the rules, requires something the model structurally does not have. To break out of a space you have to stand somewhere outside it, and the model has nowhere to stand except inside the data it was trained on. It cannot get outside its own training distribution by wanting to. The new rule, the genuinely new move, has to come from contact with something the existing data did not contain. It has to come from outside. And outside the data is exactly where the model cannot go on its own.

Forward-looking is a different direction than the machine faces

There is a simple way to feel this difference, and it is about direction.

Prediction faces backward. Everything a model does is, at heart, a very sophisticated guess about what comes next based on what came before. That is enormously useful, but notice which way it is pointing. It is extrapolating the past. Even when it sounds like it is talking about the future, it is really telling you what the past implies the future will probably look like. It is the trend line extended. It is more of the same, rendered in high resolution.

Creation faces forward, and forward is a different direction, not just a further distance along the same line. To create something genuinely new is to introduce a fact about the world that the past did not contain and could not have predicted. It is the data point that breaks the trend line rather than continuing it. You cannot get there by averaging what already happened, no matter how cleverly, because the thing you are reaching for is by definition not in the average.

This is the line I would underline twice. If all you are doing is recombining what already exists, you are not creating, you are reiterating. Reiteration can be polished and fast and impressive and still be, fundamentally, a louder echo. And the AI has made reiteration so cheap that we are about to be buried in it. The internet is already filling with confident, fluent, derivative content that is all echo and no source. To actually build something, something that moves the world forward even a little, you have to add input that was not already there. Unique information. A new measurement. A real result from a real experiment. Something true that you found out and nobody else knows yet.

That last part is not a poetic flourish. It is the practical center of the whole thing. New information is the only renewable source of genuine novelty, and the only way to get it is to go and make contact with reality, which is the one thing software cannot do for you from inside a data center.

Where new information actually comes from

So where does it come from. It comes from doing things in the world and writing down what happened. From running the experiment instead of predicting its result. From asking real customers real questions and recording the answers. From measuring the thing that has never been measured because it was too small or too local or too specific for anyone large to bother with.

That word, local, is the one I want to land on, because it is where the opportunity is hiding in plain sight. The giant models are trained on the giant, public, global pile of data. That pile is broad and shallow in exactly the places that matter most to a specific business in a specific place. It knows a million generic things about running a clinic or a trades company or a real estate practice, and almost nothing true and current about running one here, on Vancouver Island, in this market, with these costs and these customers and these seasons. That gap is not a weakness to apologize for. It is the single most valuable piece of ground available, because it is information that does not exist in any dataset anyone can buy, and the only way to get it is to go and collect it firsthand.

What we are actually building

This is the part of the work I am most excited about at Entoura, and it is the natural answer to everything above.

We are not just applying other people’s models to local businesses. Anybody can do that, and within a year or two everybody will. What we are doing underneath that is collecting information that does not currently exist, the real, firsthand, on-the-ground data about how businesses on Vancouver Island actually work, and turning it into something I think of as a Vancouver intelligence layer. Real numbers from real local operations. What things actually cost here. What quotes actually win and which ones lose. What actually moves customers in this market as opposed to what a generic model assumes moves them. The specific, current, local truth that the global models are structurally blind to.

The reason this matters so much is the flywheel underneath it. Every business we work with and every problem we solve adds a little more of this proprietary, real-world information to the layer, and the layer makes the next piece of work sharper, which makes the next business better served, which adds more information again. It compounds. And critically, it is not derivative, because we are not sampling it out of a shared model that everyone else can sample from too. We are generating it from contact with a real place that nobody else has bothered to measure properly. It is a source, not a reflection. It is the curve, not the derivative taken of it.

I want to be straight about where this is, because honesty is the whole brand and I am not going to oversell it. This is the thesis we are building toward and actively assembling, not a finished product I am going to dress up as bigger than it is today. But the direction is exactly right, and I am more sure of it the longer I work. The defensible position in the AI era is not the cleverest use of the model. It is owning a living source of information about a real corner of the world that the model has never seen and cannot get to on its own.

The obvious objection, and why it does not hold

Whenever I lay this out, someone smart asks the same fair question. Will the big AI companies not just collect the local data too, and swallow this layer the moment it looks valuable. It is the right thing to ask, and the answer is what convinces me the strategy is sound rather than scaring me off it.

The giant labs are optimized to do the opposite of local. Their entire advantage is scale, the broad and global and averaged. Going business by business in one valley on one island, building the trust to get real numbers out of a plumber and a clinic and a property manager, verifying it, keeping it current as it changes month to month, none of that scales the way their machine is built to scale. It is slow, relational, specific work that has to be done on the ground by someone who is actually here and actually known. That is precisely the kind of thing a company measured in hundreds of millions of users has no patience for, and precisely the kind of thing a focused local operation can do better than anyone on earth. The moat is not the data sitting still, which could in principle be copied. The moat is the living relationship that keeps producing fresh data, which cannot.

And there is a quieter point underneath it. The information only stays valuable because it stays current and trusted, which means it only stays valuable while someone keeps showing up to renew it. This is not a vault you fill once and guard. It is a garden you tend. The big players do not want to tend a garden in the Comox Valley. We do, and that asymmetry is the whole opening. A made-up version would say this is already a finished, defensible asset. The honest version is that it becomes defensible precisely because of how unglamorous and local and ongoing the work of building it is, which is exactly why the people who could out-resource us will not bother to.

What this means if you run a business

Strip it down and here is what I would tell you over coffee.

Your competitive moat in the next few years will not be your access to AI, because everyone will have that and it will be cheap. It will be your proprietary information and your feedback loop with reality. The data only you have, because it came out of your actual operation. The thing you know about your customers that no model was trained on. The willingness to go measure something true rather than ask a model to guess at it.

If your strategy is to point a generic model at generic data and produce generic output faster, you are in a race to the bottom with everyone else doing the identical thing, and the floor is very low. If instead you build a habit and a system for capturing the unique information only your business can generate, and you let the AI do the derivative heavy lifting on top of that source, you have something nobody can copy, because they do not have your source. That is the whole strategy, and the AI does not change it so much as it raises the stakes on it.

Where this leaves us

The cheapest thing in the world right now is a remix of what already exists. The machines have made reiteration nearly free and nearly infinite, and the next few years are going to be loud with it, a rising tide of fluent, confident, derivative noise that all sounds new and is all old.

The valuable thing, the thing that gets quietly more precious as the noise gets louder, is the opposite act. Going out into a real place, finding out something true that nobody has written down yet, and bringing it back. That is what creation actually is, under all the talk. Not rearranging the past at higher resolution, but adding a genuinely new fact to the world and building on top of it.

That is the work I am interested in. Not the biggest model, since we will all be standing in front of the same one soon enough. The freshest source. And there is a whole island out here full of information that does not exist in any dataset yet, waiting for someone to go and make it real. That is the part I cannot wait to build.