ValidationJuly 14, 2026By Maarten Follaets

Innovative AI: what actually makes it innovative

Cover reading innovative AI, showing an impressive demo on one side and a person whose behaviour actually changed on the other, with the Foxy mascot in the corner

Quick answer: Innovative AI is not the model. The models are rented by everyone, and adoption is nearly universal now: Stanford's AI Index found 78% of organisations reported using AI, up from 55% the year before. When everyone has the same engine, the engine cannot be your innovation. The innovation is the problem you point it at, and the proof is that someone changes what they do because your thing exists.

I demoed something last week that took me two evenings to build.

It was good. It read a pile of messy notes and pulled out the pattern I would have needed a full day to find myself. I was proud of it, obviously.

Then I asked myself the only question that counts. Would I open this again on a random Tuesday when I am tired and busy?

I was not sure. And an AI product you are not sure you would use yourself is not innovative. It is impressive. Those are different words for a reason.

The model is not the innovation anymore

A few years ago, having the model was the moat. Now you rent it. So does your competitor, and so does the founder who has not started yet.

That is not a complaint, it is the best thing that ever happened to small teams. But it kills a story a lot of founders are still telling themselves: that "we use AI" is the differentiator.

It is not. It is table stakes. It is the electricity, not the product.

So if the AI is not the innovation, what is?

Innovative AI is a changed behaviour, not a clever output

Here is the test I keep coming back to.

Name the person. Name what they do today. Name what they do instead, once your thing exists.

If you cannot say that in one sentence, you do not have an innovation yet. You have a capability looking for a home.

Real innovation shows up as boring evidence. Someone stops opening the spreadsheet. Someone cancels the agency. Someone does in ten minutes what used to take them a day, and they do it again the next week without being asked.

That is unglamorous, and it is the whole game.

Where the actual innovation hides

If the weights are a commodity, the value moves to the places that are not:

The problem. Choosing a pain that is real, frequent, and expensive is 80% of the work. Most founders spend 5% of their time here.
The context. What do you know or can you access that a generic model cannot? A workflow, a dataset, a relationship, a regulation nobody wants to read.
The workflow fit. The best model in the world loses to a worse one that lives where the person already works.
The trust. In anything that matters, people need to know what happens when the AI is wrong. Most demos never answer that question.

None of those four are technical problems. They are customer problems. And you cannot solve customer problems from inside your own head, which is the part I keep having to relearn.

How I test an AI idea before I build it

I have built and sold a startup before. If I am honest with myself, I was figuring a lot of it out as I went, and some of it was luck. This time I want the evidence up front, not the story afterwards.

So before I write code now, I try to do this:

Write the idea as one sentence, then underline the assumption that has to be true or the whole thing collapses. Usually it is not "can AI do this." It is "does anyone care enough to change what they do."
Talk to ten people who have the problem. Ask about last time it happened, not about my idea. The moment I pitch, the data is contaminated.
Try to take money, or something that costs them something. A pre-order, a paid pilot, a calendar slot, a real commitment. Enthusiasm is free, and free signals are worth what you pay for them.
Only then build the smallest thing that could prove me wrong.

If that sounds slow, it takes about two weeks. Building the wrong AI product takes about six months and you feel great the entire time.

If you want the longer version of that, I wrote it up in how to validate an idea without building an MVP, and the honest version of what AI does and does not change is in AI innovation is not the hard part.

The uncomfortable part

AI made building cheap. It did not make choosing easy.

We now live in a world where you can ship a beautiful product nobody needs, in a weekend, and get twenty likes for it. That is a trap dressed up as progress.

The scarce skill is not building anymore. It is deciding what deserves to exist, and being willing to hear that your favourite idea does not.

That is why I use an AI co-founder for the part where I am least reliable: judging my own idea. Foxy is built to push back, not to cheer. If you want to run your idea through that before you build it, start here.

So let me turn it around on you.

The AI thing you are building. Who changes their behaviour because it exists, and have you actually asked them?

Frequently asked questions

What is innovative AI?

Innovative AI is AI that changes what someone can actually do, not AI that looks impressive in a demo. The model is a component. The innovation is the problem you aim it at and the behaviour it changes. If nobody works differently after you ship it, you built a clever feature, not an innovation.

If everyone uses the same models, how can my AI product be innovative?

Through the problem, the data, the workflow, and the trust you earn. The weights are a commodity you rent. What is not a commodity is knowing a specific customer's pain better than anyone else, having access to their context, and fitting into the way they already work.

What are examples of genuinely innovative AI?

The pattern is always the same: a task that used to be too slow, too expensive, or impossible by hand becomes routine. Reading thousands of documents for one real answer. Drafting something you would otherwise never have written. The test is not how clever the output is, it is whether the person keeps using it on Tuesday when nobody is watching.

How do I know if my innovative AI idea is worth building?

Find the assumption the whole thing rests on. Usually it is someone has this problem badly enough to change what they do and pay for it. Then test that on real people before you build. Ten honest conversations and one attempt to take money beats three months of engineering.

Is AI making innovation easier or harder?

Both. Building is easier than it has ever been, which means building the wrong thing is also easier and faster. When everyone can ship in a weekend, the scarce skill is not building. It is choosing what deserves to exist.

Can an AI co-founder help me pick the right AI idea?

That is exactly what we built Ventropolis for. Foxy is an AI co-founder that pushes back on your customer discovery and your willingness-to-pay tests, so you get an honest read on the idea instead of a nice one.