The Work Nobody Asked Me to Do.

There is a strange part of building Atom Foundry that is difficult to explain if you look at a company mainly through revenue, customers and products.

I spend a huge amount of my time doing work that nobody asked me to do.

I am getting close to 50 studies now, and almost every one of them started with a question I asked myself. Nobody came to me and said we need this study. Nobody paid me to run it. There was no customer waiting for the answer.

I simply wanted to understand something that I could not find a good answer to.

One of the things that drives me most is understanding how AI chooses and recommends certain brands, and why one brand appears in an answer while another does not. The deeper I go, the more questions I find.

A study gives me an answer, but that answer usually creates another question from a different angle. Even before I finish one study, I am often already thinking about the next study that could add another piece to what I just discovered.

That is how the research keeps growing.

One of the best examples started with a data engineer named Rami.

We had published research about AI recommendations, and Rami challenged part of what we had found. He suggested that some of the things we could not explain might be connected to the retrieval layer, meaning what the AI finds when it searches the web.

That gave us another question to investigate.

We designed a controlled experiment where we asked the same 50 buying questions with the same model. The only thing we changed was whether web search was enabled. We ran each prompt 10 times under each condition and measured how much the recommended brands changed.

The result was 77 percent.

When web search was turned on, 77 percent of the recommended brands changed.

But that was not the end of the study. It was the beginning of another round of testing.

We kept looking for weaknesses in the result and checking whether we had interpreted the data correctly. That process eventually exposed a problem in one of our earlier claims about the difference between models. We had described a 6 percent change that had never actually been measured. When we measured it properly, the number was 66.9 percent, and a later clean comparison gave us 68.9 percent.

I was actually excited about that.

Not because being wrong is enjoyable, but because finding something wrong in your own work gives you the chance to make the work stronger. For me, research is partly about putting your own assumptions under pressure and seeing what survives.

That is also why I can spend ten days testing the same question from different angles without feeling like I am wasting time.

I want to see where the result breaks.

If it breaks, I learn something.

If it survives, I understand the result better.

Either way, I have gained something that I did not have before.

That is what keeps pulling me deeper into the research.

I am not interested in producing studies simply so I can say that Atom Foundry has published another study. I am interested in questions that are still open and areas where there is not yet a clear answer.

A lot of what I am working on today falls into that category.

I am trying to understand something that is still poorly understood, while the market around it is already starting to sell answers.

That is one of the reasons I spend roughly half of my working time on research.

I have stopped worrying about whether that is too much time because I believe the research is part of what Atom Foundry is.

If I had decided that Atom Foundry needed to make money as quickly as possible, it would probably look very different today.

I could spend much more time chasing revenue, building things that are easier to explain and selling services around ideas that already have a market.

Revenue obviously matters. I need Atom Foundry to become a real business, and I want it to become a very successful one.

But I do not want revenue to force me into pretending that I understand something before I actually do.

I would rather spend the time building the evidence first.

That matters to me because I do not want to become another company that builds a product around assumptions, puts a large price on it and then uses impressive looking charts to make the story feel certain.

I have already seen enough of that happening around AI and SEO.

I once came across a product from a very large SEO company that was being positioned around AI and GEO. When I went through the material, some of what they were saying matched what I had measured, but a significant part did not.

So I contacted them.

We went back and forth about the methodology and the assumptions behind their claims. I explained where I believed their measurements were missing important factors and why that could lead their customers toward the wrong conclusion.

They eventually acknowledged the issue. At one point they even changed their response after I challenged the original explanation.

I then shared my own methodology and the research behind it, which is publicly available through our studies.

That experience was important to me because it reinforced why I am doing this in the first place.

I do not want Atom Foundry to be built on something that sounds convincing.

I want it to be built on things that can survive being challenged.

That does not mean I think I am always right.

Actually, it is the opposite.

I expect to be wrong sometimes. I expect some studies to lead nowhere. I expect some assumptions to fall apart. I expect new evidence to force me to change my mind.

That is part of the process.

What I do not want is to sell something while knowing that the foundation underneath it might not be true.

I also do not want to discover five years from now that Atom Foundry was built around a bad assumption that I could have tested when I started.

That is probably the biggest reason I keep doing this work.

I want people to trust Atom Foundry.

Not because the company says it is right.

Not because the website looks impressive.

Not because someone created another nice looking AI report.

I want that trust to come from the fact that the work can be examined, challenged and repeated.

That is also why I publish the research.

The studies are not meant to be perfect pieces of marketing. They are part of a growing body of evidence that I am building over time.

I am getting close to 50 studies now, and I see them as pieces of the same larger puzzle.

One study answers one question.

Another study looks at the same problem from a different angle.

Another tests whether the first result survives a different condition.

Another exposes something we did not understand before.

Eventually, those pieces start connecting.

That is where Atom Foundry is heading.

I am also building something much bigger in the background that I am not ready to talk about yet. It will eventually connect the research, AI Agent Snapshot, Founder Lab and Recommendation Intelligence in a way that brings the work together.

I am not going to explain exactly what that is yet because I want to build it before I talk about it.

For now, I am still collecting the pieces.

The interesting part is that I am starting to see other people move in the same direction.

There are still very few of us working on these questions at this depth, but I am now talking to people who share some of the same ideas. Some of them are even sending me their own data and observations, and we compare what we are seeing.

That matters to me.

For a long time, much of this work existed almost entirely inside my own head and inside my own research environment. Now I am starting to see other people finding similar patterns independently.

Maybe the market is slowly catching up with the questions we have been asking.

Maybe some of the things that seem strange today will become obvious later.

I do not know yet.

What I do know is that I would rather spend the time finding out than pretend I already have the answers.

That is the work nobody asked me to do.

And I have a feeling it may become some of the most valuable work Atom Foundry ever does.

From time to time, I will share a Founder's Reality Check. Not startup advice. Not growth hacks. Just honest stories from the journey of building something from nothing.