There is a sentence most physicians are trained never to say. Three words: I don't know. Thomas Clozel learned early — from Valentin Fuster at Mount Sinai — that those three words are not a confession of failure, but the beginning of discovery.

Thomas is an oncologist, co-founder, and CEO of Owkin. He trained in Strasbourg, then crossed the Atlantic to work under Fuster in cardiology and alongside Olivier Elemento and Ari Melnick at Weill Cornell in computational biology. Ten years ago, he left clinical medicine to build a new kind of superhuman intelligence for biology — on a contrarian bet that chemistry was the confined, solvable problem, and that biology, the messy, causal, unsolved one, was where the real discovery lived.

In this conversation, we trace that arc: from federated learning across the world's hospitals, to a mesothelioma study that surfaced categories of disease no human had ever named, to Owkin Pro — an AI scientist that, in Thomas's words, is beginning to have ideas of its own. We discuss why big pharma keeps missing the AI wave, why he thinks Anthropic could be the biggest pharma in five years, why the treatment for Alzheimer's might one day come from a fifteen-year-old in a garage, and what it will actually take for him to say, someday, that he has succeeded.

Biology is too complicated for the human mind.

Right?

We still have people trying to gather the complexity of biology with a few genes and few arrows or simple bioinformatic pipelines.

But there is so many skills, there is so many modalities.

There's a real chance that EntropiqIQ will be the biggest pharma in five years because they're AGI native.

They build on rezoning models that are different from scratch and then you can pile up things.

And I really see pharma still giving a very important place in late stage clinical trials, market access, distribution, manufacturing, blah, blah.

I think the R and D is being let go to other type of companies that will be AGI native.

I feel like pharma is a bit like Marvel movies the last year.

Marvel movies are always the same.

The Avengers, 5, 6, 7, 8.

And it feels like it's the same in pharma.

We want another glyp one.

We don't have a glyph one in obesity or they're all playing the same game a little bit.

Hello, hello and welcome to Precision Signals.

I'm Sean Kozen.

There is a sentence most physicians are trained never to say.

Three words.

I don't know.

We are taught to project certainty, to name the disease, to reach for the gold standard.

But my guest today learned early from one of the great cardiologists of his generation, that those three words are not a confession of failure.

They are the beginning of discovery.

Because biology, it turns out, is too large for the human mind.

We still try to explain why a person develops cancer or Alzheimer's or dies of an ordinary bacterium using a handful of genes and a few arrows drawn on a slide.

And most of what we believe, my guest would argue, is simply wrong.

Not maliciously wrong, just too reductionist an approach.

Thomas Clozel is an oncologist and co founder and CEO of Aukin.

Years ago, he realized that the way to see biology differently was to build a new kind of superhuman intelligence.

He grew up in a French family where science was the native language.

His grandmother was among the first women physicists in France.

His parents were physicians who left the clinic to build their own medicines.

Tumak trained in Strasbourg, then crossed the Atlantic, learning cardiology under Valentin Fusterre at Mount Sinai and computational biology alongside Olivier Elemento and Ari Melnick at Cornell.

He learned to code, to run a pcr, to read a scan as if disease had left behind a grammar of its.

And then, roughly 10 years ago, he left clinical medicine entirely.

Not because he stopped loving patients, but because he wanted to change more of them than a single clinic would allow.

He founded Aukin on a contrarian bet that chemistry was the confined solvable problem and that biology, the messy, causal, unsolved one, was where the real discovery lived.

In this conversation, we trace that arc from federated learning across the world's hospitals to a mesothelioma study that found categories of disease no human had ever named, to an AI scientist that, in his words, is beginning to have ideas of its own.

We talk about the challenges of innovation in big pharma, why the treatment for Alzheimer's might one day come from a 15 year old in a garage, and what it will actually take for him to say someday that he has succeeded.

Let's step into the conversation and trace the signal beneath the.

No.

Perfect.

Hi, Tomas.

Welcome to Prejudicent Signals.

Hey sir, pleasure to be with you today.

Great to see you virtually.

I've been looking forward to this conversation.

Me too.

After all these years.

We know each other for a few years.

I know.

Finally we get to do a podcast together and jam.

You promised we're gonna jam at the end.

Right.

So thanks for joining us from France, which is where you live.

And I want to start before we go into your amazing professional accomplishments as a physician and an entrepreneur, Tomas, I want to go back to your upbringing, where you grew up and what your family environment was like.

So let's take us back to the very beginning.

Very, very French.

Grew up in a very French family as well, but a family where science and was very important and impact.

And my grandmother was one of the very first women to be a physicist in France.

That was before the war.

And there was Jean Perron, very famous Nobel Prize there.

And my parents themselves, you know, were daughters.

Both of them met at med school, but quite early on starting to really like being interesting to even bring more impact towards building their own drugs.

And so my parents, you know, were physicians, moved to create their own biotech company and actually put more than 50 drugs total on the market or something.

It's very exceptional.

But remain doctor all the time and you know, they have a new company now and always trying to figure out how they can treat patients.

So having these of impact and science.

In the family, that's amazing.

So you grew up in a scientific family.

You said your grandmother was a physicist.

So very quantitative.

Is that where your interest in data science and AI and coding come from?

Or was like this is just about.

Yeah, I mean materials coding really.

I started to really learn how to code when I was at Wake Cornell with somebody you know very well, Olivier Elemento.

And coding is a language.

And I thought I kind of at the time knew about the medical language, but I was trying to figure out other languages to speak with the data, and coding was one of them.

Physics is another one.

I think physics is probably something we don't learn enough.

You know, I was discussing with Yann Lecun recently and asking him, what is the thing you haven't learned enough at school that you would try to learn if you were going back?

And he was really talking about physics.

It's another way to interact with data.

So, yeah, I mean, I always figure I was interested in new ways to interact with data and coding was part of them.

Sure.

And it seems like our field, biology and medicine and the life sciences in general, the more we learn how to measure and compute, the more the life sciences perhaps are going to be like the physical sciences.

We can compute.

Exactly.

There is so many data we are not extracting from our ecosystem.

And the way we see, you know, even when I was an oncologist like yourself, and you've been a role model to most of us here, so thank you for that.

You're being too kind.

I wasn't seeing the patient as one.

I was seeing the patients and the cancer as many clones.

Some were going evolved towards a relapse.

Unfortunately, thermosis weren't.

And I always had this kind of vision that there was a lot more to see.

And for me, data science and AI was just a way to see things differently.

Yep.

So, you know, let's go back to your interest in medicine in general.

I guess you grew up in a medical family.

Do you think that was a major influence?

Is that why you decided to go to med school?

Yeah.

I mean, just like, you know, you, you, you, you never wonder why you, you, you, you work when you, when you're a doctor, it's so, it's so like, you know, amazing to be able to, to change life of patients.

Sometimes not, unfortunately, but it was, that was the first one, I think.

You know, I love the fact of being close to people and being able to impact them.

But I think research was also very important because is there anything more beautiful in the world to be able to change somebody's life with one very little pill?

It's amazing there's no other, like, way to impact millions of people with the same pill.

And I love also this kind of, this kind of way where you can really very rapidly change people, you know, people's life with, with very small things.

Yeah, absolutely.

So Your father, correct me if I'm wrong.

Tomas is a cardiologist, right?

Yes.

Yeah.

And what about your, your mom?

She's a pediatrician, right?

She, I think she did the five quintuplets, I don't know how to say.

In France, she was doing a lot of like, pediatric intensive care.

Interesting.

Yeah.

So, but you became an oncologist.

Where did that interest come from?

I think, you know, I think like the, the, the, my interest in oncology came, came by the, the way that it was super connected to research and, and, and multimodalities that I already, very early on I was interested in so many, so the multimodal aspects of cancer where you, you do have images to look at.

You have a lot of like, gene to study patients that have a certain, like, you know, time series of events that you can try to predict.

And, and I was always very.

Oncology was something that was amazing.

And, and, and there was so much to discover.

There's still so much to discover, especially why patients have cancer.

What is the biology of cancer.

I still, we still really bad about understanding that.

Yeah, I can, I can relate to that.

That was one of my draws to oncology because of the science of it.

It's so fascinating.

And every oncologist has to be a scientist.

There are.

What was your main motivation?

To go to like, lung?

Yeah, you know, I, I, I found my way to medicine through data science and, and computer science because that's how I got started as a child.

And you know, I wanted to be a scientist, a physician scientist.

And cancer was always the mysterious nature of cancer.

The fact that there were so many things we didn't know was the most attractive part of it.

And I was naive at the time because I thought that I can be a musician, a doctor at the same time because I was doing a lot of music.

You were.

But at the same time I was like, well, I know how to program, I'm a quant and I'm going to apply those quantitative skills in my research.

And when I was in med school, though, it was a little disappointing because I realized that the quickest way for me to lose all credibility was to mention a word like AI as an example.

And it wasn't that long ago, but AI was not looked at favorably back then.

And even data science.

And you know, I always appreciated the art of medicine.

You know, there's an art to medicine that you cannot compute, but I never appreciated the artisanal aspect of medicine.

Like, why are we reading ECG signals on a piece of paper when Even back then we could have, you know, done computation on the waveforms.

You know, years ago we could have done that.

We didn't need AI so those things were a little strange to me.

But going back to when you were in med school, which was Strasbourg, right?

Yeah.

I mean, was it what you thought it would be?

Was, were you.

Did it fulfill your expectations?

I mean, not really.

You, you.

You learn about.

You learn the language of medicine and you learn how to touch patients and to understand what they have with your hands, which is very important because I think with AI now that do a lot of things, we have to go back to learning how to examine a patient.

So in a funny way we coming back to.

Yeah, I mean it felt the human side of medicine a lot.

But I was super curious about everything.

I early on participated in a competition called igem, which is biology.

Trying always to make teams and trying to be transversal every time.

And in France it's very difficult to really do research.

You couldn't do like a double track career doing research on the side.

And so I had to find my own way.

I went to the US very early on, had no help to Harry Melnick Lab in wake on health.

And you know, I think it's changed a little bit.

But it's so important to be able to see the patients throughout different lenses and computer microscope everything.

And I think that's something missing it in Europe, which is better.

So.

Yeah.

Yeah, and I loved it.

It was my first personal professional experience.

I went to.

Yeah.

And then you start to work with a cardiologist, a renowned cardiologist.

Right.

Foster was my first rotation.

I mean first.

First I was Valentine Fuster, he's very famous cardiologist.

Was when I was at Mount Sinai.

It was my first rotation.

And what he learned me, what he taught me.

Sorry, it.

He really taught me to.

To.

To and the other students to say I don't know.

Which is, you know, the Socrates kind of a paradigm.

But it's super important because saying you don't know is what you.

What can qualify a very good doctor, not very good one and will qualify the best AI with reasoning models saying I don't know.

And I think it's very important to say that and was if you can, you have only two answer with the patients.

The right one or the I don't know one.

That was just my first experience.

I went back to make a master degree with Harry Malnick.

Years later.

And that was a very defining experience in my career.

You know, that's very interesting because that's one of the hardest things for anyone to acknowledge, especially in the professional setting, especially for physicians, because frankly, that's not.

Yeah, that's not how they train us.

And it's so refreshing to hear that.

Right.

It's genuinely very rare to have people think, I just don't know, I have no idea.

And.

And it's.

And the smartest people, however, usually do.

And the best dog.

That must have been so liberating.

It is.

I mean, we can do, we can know everything.

Especially in medicine.

Right.

When you are like third, fourth line of chemotherapy or targeted therapy, you don't know what to do anymore.

And late stage patients, you don't know.

And then you have to explore.

And it's extremely important to be open to gain more growth in knowledge and ideas.

Yeah.

So you went to Mount Sinai as part of your medical school rotation.

Is that how.

Okay, got it.

And then I heard also a very interesting story our team surfaced that let me know if it's true that one of the neurology attendings, Jean Michel Gracie, took you guys out to New York City for a walk and ask you to diagnose people based on their gait.

Yeah, yeah, he was amazing.

He's still the best neurologist I've ever met.

He's a, you know, he's a professor of neurorehab.

He's in.

And I really want to say hi to him.

He's just the best clinician you could get.

And he was, yeah, just like shaking somebody's head.

We had to discover what neurological condition he had.

We had to guess if somebody with a Parkinson condition was doing sports just the way he was working.

And yeah, he was beating us out.

And.

And he's just an amazing doctor.

You don't have people like Jean Michael Grasses in neurology anymore.

When he had to, to do it like muscle spasticity, he was taking Botox just to do exactly the right muscle.

He was the only one that always knew where exactly the needle had to be to change a tremor.

And it was like anything.

It was amazing.

He's the best guy I've met and love to keep in as well.

I mean, wow, that's fascinating.

You had a lot of amazing people around you as part of your medical school education.

So.

So that was Mount Sinai.

And then you alluded to this already.

And then you went to Cornell and started to work with our mutual colleague, Olivier Elemental.

How did that come about?

Was that also part of rotation or postdoc?

I wanted to, during my medicine I was, you know, during my residency I just wanted to come and go to the US and and learn what it was to work in the lab.

I was super interested.

And this time in the U.S. it was Ari Melnick, you know, working a lot on like computational biology and, and, and, and cancer and, and with a big wet lab, doing a lot of like.

And I met Olivia at the time.

Ari was an amazing mentor and I loved the lab experience, the wet lab experience and ultimately I wanted to stay more and, and I find a way to take care of a lab.

The one from Shahrukh Shayat was the head of.

He was the head of a small lab and the professor of URO oncology GU tumors.

And I stayed there for many a few years actually and best experience.

The US is amazing.

And this kind of part of New York, when you have Mount Sinai, sorry when you have Cornell, you have Rockefeller and msk, Sloan Kitzlering.

It's so dynamic.

You can just go and learn so many stuff and I really loved it.

Interesting.

So you did the rotation at Mount Sinai, went back to France and then you went to Cornell and that's while you were at Cornell.

It seems like that's where you hone your also quantitative skills as you.

Yeah, I learned how to code, I learned how to see medicine differently.

I learned how to use, I learned how to do pcr, have a microscope, analyzes the images, understand how NGS works, everything.

It was amazing.

And when I went back to clinic, back to France, you know, I was like so excited about the fact of being able to see things differently but I couldn't really apply it because I was so smart in the day to day clinical care that I couldn't do my research anymore.

And this is the time where I say, you know, I have to do.

I love research, I love patients, but I really want to change the world.

I have to find a way to create my own company because this is the only way I could really apply all these skills.

Like you know, alone.

So not the research anymore at the hospitals.

I decided to leave the world of clinical care to create my company.

Okin like 10 years ago.

Amazing.

So it was essentially you had learned so much during your education that the clinic seemed like wasn't enough in a way.

Right.

I was really good when things were super complicated.

They care about you know, a bit like being super organized, everything.

It's not me.

Right.

I'm a creative person and I was.

Yeah, I guess it was not made for me and.

And I couldn't express my creativity there.

So.

Yeah, I mean, I think you can when you have.

If I had to research on the side and go the back and forth, but I couldn't do both.

So it's easy to leave.

Sure.

So we're not really 10 years exactly.

A little bit less, but time flies, right?

Yep.

And what was the thesis at the time?

What were you thinking?

What did you want to accomplish with Okun?

Really?

Like when I met Gilles, my co founder, who was at the time assistant professor at a very good school called Normal Superior in France, we both had the same frustration about the fact that we needed a new intelligence for biology.

And.

And we were surprised how much shallow we were on understanding why people are sick and from cancer.

Just a common cold.

We don't even able to understand really why patients have a common cold, how people react to infections, everything.

And so we were like, we absolutely need to build the tools and the way to understand biology differently.

A lot of people starting other people at the time there was like a few companies like Atomwise starting to do kind of AI for chemistry.

But for us, chemistry was something that was not the biggest problem.

Chemistry people have been okay to do chemistry and.

And in general I think AI have a bit overfitted on doing chemistry.

But first chemistry was something that good chemists in Germany would always be good at it or in China.

But we were like, we want to understand biology.

So we decided to find a way to create a new operating system for battery.

Well, that's very.

You were in a way ahead of the curve, by the way.

OK. And I was at the FD at the time.

Yeah, at the time, yeah.

And Tomas, what I started to think about was, wow, this is such an amazing company.

Finally a company that speaks a language I understand.

And the fact that you were going after the hardest problems, which is, as you said, biology, not chemistry.

Because chemistry is a confined space and AI can learn chemistry quite well.

And we've done quite great work right now.

Protein engineering, small molecule development.

But that's not discovery.

Discovery is really learning more about biology.

The causality.

Right.

I mean we're still on causality quest.

We're not there.

But I feel like for me, the superintelligence in biology, the biology ASI as we call it, is really when we go to the cause of diseases, we can't have a world where we have no idea why patients get cancer or why people have Alzheimer.

And if we don't get there, I mean, I don't think the progress will be incremental.

We can still discover amazing things.

But I think we, I mean, the, the adventure of trying to understand the cause of biology and diseases is what excites me the most.

Yep.

In a way, you know, we can engineer using AI and other methods, the best protein or the best small molecule, but it could still fail because we don't understand the biology.

I get to target what and to target what and then to do a clinical trial in which indication which population, you still go blind still using mice models.

You know, I mean, chemistry is really important.

And chemistry is very close from the value of pharma.

Right.

Because this is the ip.

The best IP is composition of matter ip.

It's very close from the end value, but it doesn't solve the thing.

Right.

And I think that was really the bet.

But of course requires a lot of tools and we can talk about this.

Yeah, absolutely.

And how did you convince folks that working on biology using AI was the right path?

Because again, that's probably the hardest lift.

You need a lot of data, you need a lot of computational power.

How did you convince folks?

Because it was in a way, a contrarian view, especially at the time.

Right.

I mean, I think we were at the beginning we just convinced folks with the team.

Right.

I think we're a really great team.

Our first engineers were number one and two of Kaggle in Europe.

They were Kaggle grandmasters.

We had this really good mix, Michael Flander and I, between people that were.

I was a doctor, super interested in data science, he was a super talented data scientist that also studied biology at Stanford.

Were very, very complimentary and very driven.

And you know, I think we were also like, that was the team was very important.

But then I think, you know, we came up by very fast technology to acquire data.

That's important because data is the key game here.

And very fast.

We decided, and it was my idea of my co founder Gilles, to put a lot of bet on trying to be first in the technologies that will help us getting more data than anybody else.

And we came by the idea that data comes with trust.

Trust with hospitals is really important because the trust is the way to multiple years access high quality data, keeping your needle level curated and everything.

And this is why we went straight into this federated learning technology and the Biggest first step of okin was we're going to be first at technology as a how to acquire meaningful data at scale.

So we went straight there.

Very interesting.

And so your federated learning ecosystem scale relatively rapidly, Melody being the infrastructure.

Can you tell us more about that and how did you actually were able to scale fast and gain the trust of institutions?

As you said, trust is the critical factor here.

Being able to fleurite learning.

You don't take data out of the hospitals, you can remote access and remote execute framework.

So people were really happy about this everywhere in the world.

Even in Germany for example where data is very complicated, you know like data ownership, data transfer to cloud and everything.

So it was a very, it was a great technology because people were understanding it could break research silos and competitive silos because you know, people are not willing to share.

So it was a great way to have people working together and it was a good way to protect no rights and privacy.

So it worked out very well and very fast.

We could gain multiple year contracts with hospitals.

What was interesting is not only we had this trust, but they accepted that we were able to put our own infrastructure, servers or cloud within every hospital which gave us a lot of barrier to entry and working with the top one hospitals data that is very high quality, sending for one and generating to curate and being able to show value.

Which means writing science paper, discovering things very early on put us in a good loop of people liking our technology more and more.

And this is really how it's.

Yeah.

One of the papers that you guys published around that time was on mesothelioma.

That stands out to me.

The reason being this was digital histopathology.

When you start to work on digital histopathology at the time, a lot of the efforts we're trying to predict, for example and diagnose non small cell lung cancer or small cell lung cancer.

Basically nomenclature that we already know.

And we already know that nomenclature.

For example non small cell lung cancer is a linguistic term.

We just made it up because we looked at the biospecimen under the microscope.

We saw that the cells were.

There were two types of cells small.

So we call this small.

We want to call the other category large because it's not varying sizes.

So we just made up a phrase non small.

But and a lot of the AI efforts at the time, again I was at the FDA at that time were trying to essentially replicate our bad behaviors human linguistics.

Whereas okin, with your mesothelioma paper, you identified histological categories that have no real world correlate and it was just the AI that was able to identify.

So that to me was very groundbreaking because even today a lot of folks are uncomfortable with these novel digital biomarkers.

Can you walk us through that and tell us why?

You know, tell us about that journey of really using AI and, and the computational capabilities that you have to discover de novo realities, I. E. Biomarkers?

Yeah, I mean, I feel that, you know, biology is too complicated for the human mind.

Right.

We still have people trying to gather the complexity of biology with a few genes and few arrows or simple bioinformatic pipelines.

But there is so many skills, there is so many modalities, there's so many.

How do you go from a gene to protein, protein to cell, cell to an organ, organ to body, whatever.

It's so complex the human mind cannot cope with it.

So I think we have first to admit that everything we've discovered with the human mind, most of the things are wrong.

You know, we're surprised every day by medicine and biology.

I was discussing with a friend of mine, that is the head of intensive care in Lausanne in Switzerland, a brilliant guy, Jean Daniel Shish.

And he was telling me about, you know, this kind of new trend of seeing very young patients dying of very simple pneumococcus, very simple bacteria, sensitive to most of the things.

And we have no idea if the reaction, immune system reaction of this young persons are.

Is too heavy and we need to immunosuppress or too low and we need kind of immuno boosting.

We have no, we don't get that.

Just the way we characterize bacteria is not enough.

Probably wrong.

The way we understand sepsis, probably wrong.

The way we understand and we approach cancer wrong.

I mean, everything is wrong in a way and we need to be unsupervised in the way we.

We need to create an intelligence that can capture new signals and think by itself.

The idea of working today is really about this reasoning in biology that has to discover things with a new mind, which is an artificial one, it will be confronted to real people.

But I just think that the human intelligence won't cope with biology anytime.

And we need to be able to be open about reviewing everything we do.

But we shouldn't talk about liver cancer, bladder cancer.

We just talk about the mutations of cancer, as in the Fabrice, as Andrea's publication.

I mean, everything has to be.

We need to break our lines and we need to be open to changing the way we see things.

Right.

Do you think we're getting there or do you think we're sort of stuck in what we call gold standards?

We're still a bit stuck in Gallson.

I think we would get there when AI end to end will have bring a true therapeutic solution for places where there is nothing like Alzheimer.

I mean today we're accepting drugs in Alzheimer where there is a mild difference in mild impairment patients.

This is so shit, right?

It does work because the business works that way.

But this is not satisfying, Right.

Sometimes you can get a drug approved if you change the overall survival of patients for a few months.

Even if the quality usually change, it's not even better.

Quality of life can be even worse.

I mean we have to say we failed and we.

Let's go back.

I think people are still stuck in this gold standards.

The pharma is pushing very hard.

I think pharma.

We can talk about it.

I think it's.

Pharma has gone to an end.

Pharma.

I don't think pharma.

Pharma is still a very amazing, innovative player.

But I don't think they're able to find new things.

And they've been playing out with the same space of molecules, but there's way too much legacy.

Now of course we go V drugs are amazing.

ADC are changing life of patients.

But we could go so much further.

That's interesting.

You know, I don't think medicine was always like that.

You know, I think this whole concept of gold standards is relatively new.

I would say like in the past, like three, four decades.

We, we actually looked at progress through fighting and questioning gold standards.

In a way there are no gold standards because if you are still relying on what you call gold standard and those gold standards don't change, then there's really no true progress.

That's true.

I agree.

In a way the FDA these days doing quite a good job.

You know, they try to say let's make more real world proof of a drug that would prove being safe.

Put it on the people and see how it works.

They're kind of like breaking good.

So bad or a way of thinking there.

That's good.

But I agree with you, like a bit like that.

But I think in the last years, not so much.

Yeah, I agree.

I totally agree.

Do you think small biotech is being more experimental and more bold and venturesome?

Yeah, I think so.

More venturesome, more trifle.

But I genuinely have a belief that the future of these big Companies will be AGI native company like us, like Anthropic, like bigger ones, like anthropic.

Of course, OpenAI.

I think there's a real chance that Entropic will be the biggest pharma in five years because they're AGI native.

They build on rezoning models that are different from from scratch and then you can pile up things.

And I really see pharma still giving a very important place in late stage clinical trials, market access, distribution, manufacturing, blah, blah.

I think the R and D is being let go to other type of companies that will be AGI native.

And maybe the treatment of Alzheimer won't come by Pfizer, but it will come by a dropout that is 15 years old in his garage.

And it's not.

I mean of course you need validation, you need things, but you will have dark lab when you can connect and play around.

And if we create the right sandbox, this is where we can really win because I mean we love it.

I totally agree with you, Tomas.

This reminds me of a NMIT professor that you may know.

Andrew Lowe several years ago did research on that very theme that you talked about that big by pharma.

Big Pharma, let's call it is more has become more of a capital allocator and distribution channel.

And he actually made a direct comparison quantitatively to what happened to Hollywood that how these major movie studios, there was a time that they actually wrote the script that were very bold and experimental like back in the 60s and the 70s.

And then they became giants with a lot of capital and they started to basically become distribution channels.

You know, they purchased everything, they acquired their innovation and then they distributed through the network versus coming up with their own ideas.

And in a way most some biopharma companies have gone that way.

So do you think?

I like the analogy.

Yeah.

It was very refreshing when he talked actually I invited him to the FDA several years ago and that was one of the things that he talked about and it made sense.

So do you think.

But you know, it's very hard for small companies, biotech companies to get the capital they need to be really experimental.

They still have to follow what the investors think is the right thing to do.

And that is usually quite.

These are quite safe bets.

Or am I wrong?

Do you think there are investors and biotech companies that are really pushing the envelope without naming names?

You don't have to name any names.

I feel like pharma is a bit like Marvel movies the last year.

Marvel movies are always the same.

The Avenger 5, 6, 7, 8.

And it feels like it's the same in pharma.

We want another gleep one, we don't have a blip one in obesity or they're all playing the same game a little bit.

But on the venture side, I think some venture capitalists in biotech for example are very, very good.

Right.

If you take the Atlas Venture people have to world have new amazing knowledge and do take risks.

I think that on the other side of the tech investors biology is a more complicated story.

Biology is complicated for them as well with a bit of a longer timelines and I think there is still a gap to create the p venture capital that will really have a good vision about what is this AI company in biology can do and what are the intelligible assets and the valuation markup we can do today.

It's not even a sovereign play biology.

When you talk to, I was with President Macron and you know, at one of the dinners and he's literally like they never talk about biology and health for a sovereign thing.

Right.

But I kind of like if a population is healthy, it's the most productivity gain you can have.

Right.

And Vigov brought more productivity gain because people don't die before 65 years old than any AI.

And it's very sovereign play.

It should be thought this way.

But people are really about sovereign player on LLM cloud compute, not about health.

It's a weird thing, I don't understand why.

Yeah, no, you're, you're absolutely right.

You know, that's that sort of start.

You know, I'm thinking now about the fact that maybe in five, 10 years, as we automate the mechanics of various enterprises, let's say finance and to a certain extent all the back office stuff in medicine, then the cost of production across most sectors is going to come down quite substantially.

Then what else is there to conquer?

It's human health, it's biology, it's everything.

And even like, you know.

Yeah, I mean even like the billionaires in this world.

There's a lot of billionaires in France, I don't care until they're eight years old and like, oh my God, I really care about it.

And I think like health is the biggest bets.

Everybody should invest in health.

And you know, I think in the US it's a bit different because people are paying for their healthcare.

There's more value in Europe.

Health has no value, it's free.

You cannot speak about money and health.

I mean there's a bit of a structural problem in the other side of the world around this.

Right?

Right.

Yeah.

It's not always about developing a new drug.

Sometimes it's about reconditioning your environment.

So Tomas so Okun developed massive infrastructure, became a unicorn and accumulated a tremendous amount of data.

So now you had the infrastructure, you had a lot of biopharma clients, you have this massive infrastructure and then now what did you decide to do with all the data?

No, what we did in the last year is we had so many, even the last two years we had so many data everywhere.

That's great.

We put one infrastructure to access all the data.

That was okay.

Then we had a lot of different tools everywhere.

We had a tool to analyze pathology images, a tool to build the B specifics.

We had tools to improve clinical trials.

We had different things but nothing was really like structured into one architecture.

What happened a few months ago is I you know I have a new co CEO.

His name is Pascal Van Berger.

He's a, a young genius, 28 years old, built like five amazing companies and very software minded and we came back together and said we're going to build one infrastructure, one architecture that can cross talk between a lot of data from all over the world.

Tools skills LLM always build on cloud coding agent to build workflow of research and really figure out if we can build a resulting model.

Right.

And I think what happened for okay now is in the last few weeks this is kind of one architecture it's called looking pro is becoming an AI scientist having his own ideas, having his own or in the workflow of research and goals of discoveries and actually has found things that I think are trend mostly amazing.

So I, I think I'm.

I'm owning the only real reasoning based on real patient data model worldwide and because I'm not trained on the worldwide web like entropic would be or pan which is amazing companies and because we can validate things in the lab we can we have the right structure.

I think we're very close to bringing something amazing and and I've tested this AI scientists in research mode find me a new molecule, find me a new target but also on the clinical mode fourth line of chemotherapy for bladder cancer patients.

Can you find me new insights and I've been extremely satisfying in our network of kol but what we are discovering and I think this is most exciting time of my last maybe 10 years where I figure out we are getting closer to a new intelligence and and, and that can be way more powerful than all of us together.

That's amazing.

So okin how would you characterize okin today, is it a company that's building infrastructure or it's now a drug discovery venture?

I don't have to, I mean for me it's just the company building the future of biology AI within an AI scientist software.

The drug discovery is a no brainer because it's the ultimate product is to build solutions for patients so reasoning models without the capabilities to show first that it's true.

It's called a reality check.

Everybody should have it because.

And the second one is to push your own discoveries within diagnostics.

Actually we spin out to diagnostic company or therapeutics.

So yes, discovery will be part of and key within okin.

But you know the first is being and we have amazing products, we're in phase one, we have three more coming, we have two diagnostic tools that are approved but for me the qu it's not the number of pipeline we have, it's not the number of programs, it's really the quality of unmet needs we solve.

Right, Absolutely.

Makes perfect sense.

And so you're doing a tremendous amount of very interesting work internally as part of your discovery pipeline.

Are you also thinking about partnering with others?

Is that part of the strategy?

For example, if for a biopharma company can they reach out to Okin for discovery parties?

Of course.

And we are announcing two big deals with pharma very very soon where they take our software for different time of the life cycle of the drugs because from finding a target to chemistry, clinical trials and even competitive intelligence can be used for different things and bring huge value.

So we are working with pharma and we have a lot of partnerships.

This is interesting for us, we learn it's good for our Runway.

But I still have always this impression however that pharma isn't super clear about exactly their agentic strategy and what they really want to achieve.

And I'm just working with them knowing I want to replace them.

I think the CEO of Sanofi, Paulo Hodson, I told him this once, we're going to replace you one day or us or another AGI native company and he find it really funny and he has a sense of humor.

Yeah, let's see.

Right, right.

Well he made a huge investment in Oken, right?

Yeah, he did invest.

He was really, really great.

Right.

Whatever you told him, it worked.

So once again, no AI company that hasn't actually solved an unmet need where the pharma have felt can be considered as winning anything for now, if you ask me, do we succeed at okin and no, we have failed.

We have not Treated cancer.

The day we will, I will have succeed.

I won't have succeed because I raised a billion.

If I do, I won't have succeed because I didn't exit.

I will only succeed because I'm a doctor like you.

If I treat something that is important for the world and for patients.

That's amazing, Thomas.

And you know what you're describing basically is you're bilingual.

You know, you understand the unmet needs of patients because you've seen it, you've touched it, you've felt it, but you also quantitative in your thinking.

So you understand the language of AI and data science but you also understand the language of medicine and biomedical research.

It's a very rare combination to have in an individual.

Same.

Same, same.

Same.

Yeah, exactly.

Sure.

So do you think, why do you think it's so hard for biopharma to move in that direction?

Big Pharma?

Because Big Pharma has talked about AI for some time now and most experiments, people argue that having gone as expected.

How do you see today's biopharma AI efforts?

It's such a good question.

I think there is a few reasons and you will tell me as well because you had a great experience at J and J.

But I think first there's a cultural mismatch.

AI companies are, people are wearing their like HAL shirts, drinking cocktail, I mean I like playing video games.

It's very, very, it's not.

And pharma is very corporate still.

That's the first one.

I think the way of working is very different.

Weekly sprints for, for AI companies with like yearly sprints, it's a different timeline then I think the pharma, you know, really try to, to, to, to, to.

To integrate this AI within their pipeline, their workflow where I think they should just have a separated company building their own AI driven pipeline on the side.

A little bit like Aviv Regev doing this at Genentech where she's responsible for the pipeline which is I think works better than trying to integrate, merge, change the workflow.

Then I think they hired chief Digital officer which was a big problem.

They had people, not the best minds in AI in the world, but mostly people from BCG or whatever.

And it just doesn't work.

You cannot understand what is the predictions, you can understand what's the feel of coding if you're great.

And why, why have these people from you know, kind of bcg, whatever became like officer because they speak the same language than the CEO and the CEO.

And finally, and it didn't fail this, this chief digital Officer, sorry had two responsibilities.

One was digital apps, digital transformation.

The other one was AI.

It's very different.

Finally the budget.

You can't build an AI transformation, an agent transformation without hundreds of millions of budgets.

If I was a farmer, I would put a billion if I had to on this.

And sometimes, you know, I know Big Pharma that put a budget of 70 million, 100 million but they still say they want to be that.

And I think the last problem I want to talk about is a structural problem with the CEO of Big Pharma.

What happens with the CEO of Big Pharma?

They have such a huge salary that most of their goal is to make another year and they don't see long term.

They want very direct pipeline value.

They want to be able to stay because the stocks do good.

But you have to be transformational in how you build AI.

You need to build data lakes.

Very well done.

And to build the step by step.

And the problem of this CEO is they don't see the yearly.

They speak a lot about AI but they don't have the long term vision.

That's a bit my mission.

Sorry, it's maybe a bit wrong.

No, it makes sense.

It makes perfect sense.

I totally agree Tomas.

I think it's cultural at multiple levels folks within companies and also the markets.

The investors, the public investors that are looking at the expectations they have from the CEO.

We don't have those same expectations in the private market from let's say OpenAI or Anthropic.

Those investors are betting on those companies for the long game.

Whether you know that's a right bet or not.

We need to have the same philosophy in a way for biomedicine.

Why do you think there is the investors in.

In.

In biomedicine are so conservative and they're not long term thinkers.

Both on the retail public markets but also when it comes to private capital.

Is it.

It is really maybe a problem of the capital allocators and how they view these companies.

Maybe it's a brain of the the AI companies that hasn't in bio.

I mean once again none of us has proven to treat something completely different.

I mean like Daphne at Institute is making amazing things on ALS and we so I'm so excited to see if it works and I hope so in silico and Alex great deal yesterday had a lot of molecules that are race.

You know really, really thinking.

I think we have things amazing in cancer as well and soon in rejuvenation.

I think in the end, you know we haven't proved we haven't treated something new in 10 years, so maybe it's our fault as well.

We haven't brought the right proof points.

So I wouldn't put everything on the venture capital.

I think there's a bit of this problem, but it's a bit changing.

People speak more about AI for biology.

Even like Larry Ellison speaks a lot about it.

Obviously Dario from Anthropic, an amazing smart guy, loves biology, has studied biology so he's also very driven.

A great change and I think it will change and I think in few years it will be the hottest topic but it will always be a biggest barrier to entry because understanding biology is a little more complicated.

You need curiosity and it will.

Right?

No, it is really one of the greatest challenges of human beings moving forward again because we're automating everything else.

But biology is going to be what consumes most of our time probably in 10, 20 years.

Speaking of the REO and anthropic, you know I was playing around with Claude the other day and doing some MCP connections and then all of a sudden I saw Okin was like, wow, one of the.

Okin is one of the.

How did that come about?

It was just one.

Yeah, we just went up the connector and actually we built our Okin key on cloud coding agent.

Yeah, Anthropic is amazing.

Right.

The way they think step by step what they're building is amazing.

But you know, solving coding is a bit like chemistry.

You can go back the algorithm biology is a bit more complicated but if there's what to succeed might be anthropic.

I think I'm very admirative of the company.

So going back to the connector, the Anthropic connector.

So is that for folks that are working with okin they can use Claude to connect to the data assets.

We have a lot of tools connected by MCP but our workflow is built on cloud coding agents.

So we always code new workflow of research, we can build campaigns and this is just the way we built.

So we built our platform orchestrated and making the tool coding by cloud.

Right.

This is how it works and it's amazing.

Of course we have to pay entropic every run.

Right.

Limitation in the business model and maybe, maybe better in China cheaper but for now.

Right, right.

Well, what do you think about that?

Actually, you know these companies, you're right these to make these API calls to OpenAI anthropic, very expensive but local models and a lot of them being open source models are getting quite good and our computers are getting faster and faster.

You know, I just got the latest MacBook Pro that has a neural engine for the first time.

So I'm running local models and I'm surprised how fast it's actually handling it.

Where do you see is that, do you think where the world is going?

Where maybe in 10 years nobody's tapping into the cloud, we're just running our own local models maybe.

I mean we did our own mid size models, Kurosokin zero which is still being called by our platform and it really gave us a great sense of and feel of data because we fine tune it with multiple like reinforcement learning cycles, our patient data, our lab data.

It was great.

It still costs money, right on the compute side I think however, when you'll be able to train large models on the cloud like Miramurati company for cheap, it will be a game changer.

So I think that's driving a lot of differences as of today.

Still too expensive.

And it's a big bet we took it because we have so many amazing Data sets and Okin 0 outperformed like Gemini and Claude on many, many different levels.

Still it costs.

Yeah, yeah, absolutely.

There's this very well known music producer, he's been around for a long time.

His name is Rick Beato.

He has his own YouTube channel.

He was on Lex Friedman a couple weeks ago.

So he's making very interesting correlations between the music industry and the world of AI.

So he talks about the fact that not long ago you had to pay thousands of dollars an hour to record your music in a music studio.

And he likens that to sort of how we're now tapping into Anthropic and OpenAI and Gemini and it's very expensive, these tokens are very expensive.

But then something changed.

You know, everyone started to have a home studio because our desktop got so powerful and Pro Tools and Logic Pro really changed the game.

So everyone started to build these home studios and these very expensive studios essentially almost overnight disappeared.

You know, they couldn't basically manage, you know, they were, it was so expensive to, to run those enterprises, brick and mortar enterprises.

I mean I'm sitting in my own home studio.

You know, this home studio would have cost, you know, right now it's not that expensive to have a, have a home studio.

But 20 years ago this would have cost me millions and millions, you know, outside of the reach of most consumers.

So here his thinking is that that's what may happen to these LLM giants as the local running your own local models become much more doable for the consumer.

The Average consumer.

But we'll see, we'll see where that goes 100%.

But I think I know the move of acquiring the founder of openflow for OpenAI just gonna get OpenAI closer and closer from the homes because people are going to use it even more towards their WhatsApp workflow and everything.

And maybe this is what they have in mind is the individual homes are where we need to do.

I mean, I love that individual homes will be the best on type of biology and people in the lab will have their kind of connected lab somewhere in the city and just make the little experiments, try it on the lab in the city, robotized lab and find something every day would be really fun.

Maybe there.

Well, yeah, I mean, maybe that's right.

I think what he was referring to.

Right.

What was he referring to?

Was.

As well.

Right.

Maybe the homes will be.

I mean, I love to have dropouts trying.

Yeah, right.

That's true.

Right.

Well, I don't think we're going to get there.

I think what he was referring to, Tomas, was just running your own LLMs.

I think you still need lab, you need a wet lab, you need that expertise.

You need to combine, you need data.

You know, I think you worked for example, so hard on creating this massive, amazing federated data ecosystem.

I think that is the fuel that we'll want here.

People are doing lab on the chip, you know, why not seeing Apple stores in five years, lab on the chip, build your own little lab, trying to find your own treatment if you're sick with your own little LLM you plugged in.

Yeah, I don't know.

And a great company called Blossom I invested in and the CEO Yan Flow is amazingly smart, right.

And he can do high throughput kind of microfluidics things.

And maybe the last, you know, I would love that.

That would be the best future ever to have that people playing around.

Interesting.

So unfortunately we're almost out of time, but I want to ask one last question, speaking of the future, Tomas.

So based on where you see things are going, what do you think the world in Dubai, medical enterprise and drug discovery is going to be five and ten years from now?

Specifically what will be able to do in five, 10 years that we're just simply not capable of doing today?

I really think that it will be once again five to 10 years, people in their home discovering drugs.

Because I really think that in 10 years I will not.

In five there will be this like robotized lab will work better than today.

It will be commoditized and people will be able to build their own personalized treatment.

Because the future of, I don't know, ad sync cancer is that everybody could build their own.

Medicine will be truly personalized.

I think in 10 years, yeah, medicine will be way more personalized.

AI will help matching new biology to the right people.

I think we will discover.

Maybe we will discover more ways to approach real curative options for cancer, Alzheimer's.

But, you know, 10 years, I hope we will be there.

I'm not sure we will.

I still believe that pharma will change.

It will adopt AI strategies for R and D. Fully R and D agents will replace most of the R and D. Pharma will be much smaller companies.

That's it.

And yeah, you know, I think like, we.

I hope we'll.

We'll develop a new intelligence that will do that.

But I think we're still missing some key data that's about viruses.

We know 4% of viruses that are about environments.

And.

And hopefully the data generation will match the need of TCI revolution medicine.

Right.

So that's a very promising future.

But it seems like we need to be very focused on having data liquidity, basically, because it's all about the data.

And that could be the bottleneck.

It could be the bottleneck.

Tomas, thank you so much for your time.

This was a fantastic conversation.

No, it was great, Sam.