It seems almost obvious to me that you'll have large-scale automated factories that are running experiments at a scale beyond anything that we can imagine today.
The pain point really is around intellectual property, single supplier risk, and geopolitical risk. There are now multiple documented examples where key intellectual property is getting leaked out of these CROs. As a biotech, you are really giving your crown jewel IP, and suddenly all of your competitive advantage is lost. The labor is much more expensive in the US, and so typical price points of US CROs are kind of four to fivefold higher than the offshore CROs.
How do you thread that needle, and how do you build on top of that?
There's a lot of thought that goes into it. Things happen that are unexpected. One of the benefits for us is...
Hi everyone, I'm Guru Chahel, and this is The Investment Memo — the show where we walk founders through the actual memo we wrote before we invested. Today we're sitting down with Kevin Weil and Saul Riceberg, two of the co-founders of Warp Sciences. Kevin and I go back to the very beginning of our careers. He's taken early technical products to enormous scale at Instagram, at Twitter, and most recently at OpenAI, where he led the product team. Saul has taken five drugs into clinic and has years of experience that offer an incredible view into the space.
Science today is bottlenecked by how many experiments humans can run, and AI itself is bottlenecked by the scientific data coming from those experiments. Warp is building a robotics-first research lab right here in the US, where experiments are both a customer deliverable and a training sample for AI to get better at running those experiments. Kevin, Saul — welcome to the show.
Kevin, I'd love to start with just an introduction. You've had such an incredible career across multiple domains. Just share a little bit about your journey and what brought you to this point.
I grew up in Seattle, and my dad was at Microsoft for years and years. So I grew up programming when I was a kid, but then got really into math and physics. Came out here to do a PhD in physics. I thought I was going to be a physics professor for the rest of my life, until my wife Elizabeth, who was much more involved in the entrepreneurial ecosystem at Stanford, kind of introduced me to the idea that there are all these startups. And rather than doing, you know, math in a room for a while — which I love, to be clear — I could go write some code and ship something to a million people tomorrow, and it just felt so much more tangible.
So I dropped out of my physics PhD, started working at startups, including the one where you and I worked together at Tropos, like a million years ago, and then went to Twitter in the early days as a data scientist engineer. Switched into product, kind of accidentally ended up as head of product at Twitter, over about seven years there. Went to Instagram as head of product. Helped start the Libra crypto project inside Meta. Went to Planet, where we were building satellites that image the whole world every day, and then to OpenAI as chief product officer, and then ultimately helping start the science team there.
That's incredible. I mean, just by my account, that was network infrastructure, consumer products at scale, financial services and the crypto space, and then the frontiers of AI.
Yeah. I mean, look, I think personally I love half knowing what I'm doing so that I'm not totally useless, but half having no idea what I'm doing. I think people generally underweight their ability to learn new things. And a big part of it is getting to work with amazing people who can teach me. I learn from Saul every single day, as well as the rest of my co-founders and colleagues, and it's a privilege.
I love feeling like I'm constantly having to keep up with incredible people that I get to work with.
Yeah. And how about you? What brought you to this place? Is it as diversified and as varied as Kevin, or were you much more focused?
Yeah, I think my path has been much more on the depth side and much less on the breadth. I have spent the last several decades in life sciences. Got my start at Gilead Sciences, the big pharma, working on immuno-oncology projects. Realized that bringing medicines to patients is really what drives me, and my focus in career has been sort of laser-locked on that goal from then on.
Much as I love Gilead, I realized that things can move faster in small company environments, and so I moved to several small companies, including Schrodinger and also Terramto Biosciences, most recently where I was leading discovery and put several drugs into the clinic, for both oncology, for cancer patients, and rare hematological disease. Throughout, I think I have found this deep love and passion for bringing medicines to patients. But I also have found this frustration, where the velocity of bringing medicines to patients is not really accelerating. And I watch peers on the AI side, on many of the other tech verticals, kind of hyperscaling the rate of growth — kind of the second derivative of how fields move forward. And I just don't feel like the same thing is happening in the life sciences yet. And so I really wanted to build a company where we could change the velocity of discovery rather than just making incremental discoveries.
And then Kevin, talk to us a little bit about the last project you were leading at OpenAI, and then what led from there to here. And as part of that, maybe we can introduce your co-founders in addition to Saul as well, just to give the audience a little bit of a preview into how did the team come together and what is the mission of the company.
Yeah, so when we started working on science within OpenAI, we started with mathematics and physics, and we were especially focused on mathematics, I'd say for a few reasons. One, you can do it all in silico, which is nice, right? You can just throw more compute at the problem. And we had a number of incredible mathematicians on the team, which is very useful when you're looking at things like synthetic data and creating RL environments that are custom for the kinds of math that we're trying to push the model on, because especially when the models get to the edge of human ability, you need incredible mathematicians to judge whether what's going into the model is really going to push it forward. It's really cool to see how that has all gone now, with Navier-Stokes being solved and all these other things. Models are definitely superhuman in mathematics, and I think there's a lot to be excited about there.
But when you start to think about the physical sciences, they're completely different. Rather than being theoretical and being able to do them all in silico — biology, material science — they're all experimentally driven. You move the field forward with data and experiment. And I think the reason that we see superhuman models with mathematics, and obviously software engineering and these other things, but you don't yet see superhuman AI models for chemistry, biology, material science — like, we're starting to see cool things, but not yet what we're seeing in mathematics — is because the models are data poor in the physical sciences. They just don't see all the experimental data that you would want them to see in order for them to learn everything, then begin to make more rapid discoveries.
And if you believe that, then the way to solve this problem is to generate data at scale. And I think to do it in a focused way: science is a really big place, so if your mandate is so broad that you generate data as sort of a light dusting of data across all of science, it's going to be hard to make real forward progress. And so you start thinking about where is it scientifically relevant, commercially relevant, geopolitically relevant to generate data in a focused area at large scale. You start thinking about autonomous labs, things like that. That was what I was excited to work on as I left OpenAI.
Saul and I got connected, and Saul was coming from his background and was thinking about, well, this big chunk of drug discovery is lead optimization, is working with a CRO. All of this has been outsourced to China, which is an IP risk. It's a geopolitical risk. There are real problems here. Basically, what we've done in manufacturing, we're beginning to do as a country with biotech and pharma, except it's not too late. We can bring it back, but what will it take to bring it back? We need automation and robotics, and a lot of the same things that I was just talking about, about how you generate data at scale. So in some ways these problems come together, and there's an opportunity to reshore the clinical, or the preclinical small molecule CRO, to do it with robotics and automation, to do it at large scale. There's a huge industry that relies on this. We want to bring it back to America. We want to do it in a better way, and then there's an opportunity to build really incredible AI models as part of this as we do it.
It's an incredible mission. Kevin, what expertise do you need, and how did you pull together the founding team to be able to execute on this mission?
Saul and I got connected and kind of instantly realized that we were thinking about different aspects of a similar problem, which was very exciting. Two other co-founders: one, D.J. Patil, who has been a data scientist going back so far that he actually was one of the inventors of the term data science. He was early at LinkedIn, and then co-founder at Relate IQ, now Devoted Health, had a stint as the US chief data scientist in the White House. Just an incredible human being, somebody that I've known for 20-plus years, somebody that I've always wanted to work with, and now we have the chance.
And then my other co-founder, our other co-founder Mono, went from intern to VP at Meta. So he was one of the co-founders of FAIR, Facebook AI Research, and has been building AI models like Llama and Muse at Meta Superintelligence Labs for the last bunch of years.
That story is actually a really interesting one. As we were getting started, we were just, you know, very early exploring ideas. We had a sense of what we were going to do. We met these two roboticists, and at the time we're literally working out of my wife's little... my wife runs a venture firm called Scribble Ventures, and she has a little office in Portola Valley. We were working out of this tiny little office space.
And we met these two roboticists, Sarthik and Shagon, who had started a company. They were very early as well. They'd raised a seed round, and we realized they weren't doing exactly what we were doing, but they were close enough. And we were like, you know, why don't we just join forces? And they got excited about what we were doing. We really liked them. And so, as part of joining forces, they were going to give back most of the capital that they raised in their seed to their seed investors, and we welcomed their investors into our round as part of that. One of their investors was Mono, and when they went and told him what they were going to do, he said, 'Well, wait a minute, that sounds really interesting.' And so we got in touch, and that's how Mono became a co-founder.
That's incredible. That's how the team of four comes together with expertise in AI, robotics, chemistry and life sciences, as well as data science.
And it looks like the perfect combination.
I feel very lucky to have the co-founders that I do.
And I got to say, Kevin, I see the common thread in the values alignment in the four co-founders. It's incredible. I think it's so important in building the right kind of culture. So I'm excited that the team came together in this way. Now you are taking on a very complex topic though, before we dig into how you're solving this problem. So since you've been, you know, living and breathing in this space, tell us about this market: what is a CRO? As a chemist, why are you even interacting? What is this thing?
Yeah, maybe the best way to start this is to just zoom 10,000 feet above and talk about how does drug discovery even work on the small molecule side, and how do we use CROs. If we think about a modern small molecule biotech, generally any given company is founded with some molecule in mind that has some hint of biological activity. Maybe it binds to a protein of interest very, very weakly. But generally these molecules are not acceptable as a drug. They might not be soluble enough. They might not be potent enough. They might be toxic. There's some set of problems that make them not a drug, even if they sort of have hints of being drug-like.
The process of running a biotech is almost a directed evolution, or reinforcement learning, of this molecule. And so the idea here is, on day one of the biotech, you manufacture 50 variants of that first molecule hit that you had, and each of those 50 variants is a powder in a vial. You test each of them for solubility and toxicity and how potent they are, etc., etc. And maybe one of them is slightly better. And so you walk up the gradient. On week two, you make 50 variants of that slightly better one, right? And so it really is just like directed evolution, where you are kind of survival of the fittest molecules, and walking up a gradient to the molecule that is eventually a drug to enter clinical trials. Now all of the physical work that I just described is, in the modern era, outsourced to contract research organizations, or CROs. So the job of the biotech company itself is to say, 'This week, here are the 50 molecules I want made and tested.' That order gets shipped out to a company, generally in China these days. They make those 50 molecules. They test those 50 molecules on your behalf for how potent, how soluble, how toxic are they, and then they send you that data back. And your job the next week is to say, 'Okay, I analyzed all of this data, and I decided here's the next 50 that I need to make.'
Yeah. And so you can see the CRO is sort of, in many ways, the creator of all of the physical data and physical entities that occur in this process, right? They are making the molecules and they are testing the molecules. Often times the biotech itself will never have laid hands on the drug until they're running in the clinic.
So, it seems like a fairly mature industry. You've got CROs out in China, labor-intensive job, lots of mixing, running chemical experiments. Where's the pain point? What's wrong with this industry?
So the pain point really is around intellectual property, single supplier risk and geopolitical risk. So I'm really of two minds with the existing CROs. There are really two major players in this space. One is called WuXi and one is called Pharmaron. From a services perspective, they are both fantastic, right? They make the molecules you want with very high quality, with good speed. They're affordable. But the trade-off of this is, first of all, single supplier risk is always scary, right? Having only one or two companies who can do the thing that your company absolutely relies on is always an issue. And that single supplier risk becomes even scarier when both of these companies are based offshore, and given geopolitical tensions, this can be a very kind of risky endeavor.
The final piece here is, there are now multiple documented examples where either inadvertently or maliciously key intellectual property is getting leaked out of these CROs. And so there are examples where, as a biotech, you are really giving your crown jewel IP — this is the molecule that I think could be the drug — to the CRO. They have the ability very quickly, there's no reverse engineering needed, even just to export that, to build their own competing company, and suddenly all of your competitive advantage is lost.
Because they are the hands and brain for doing five years of experimentation on your behalf. And so when you get to that end point, they have access to the same IP that the biotech here in Silicon Valley might have had. And now you've got this huge problem: the five years of work that you put in, and that you actually paid somebody else to do, they have access to the same thing that you did.
And as Kevin alluded to, we have seen analogues of this on the tech side, right? The Apple and Foxconn story, I think, is a very parallel challenge in a lot of ways, where you outsource more and more and suddenly realize huge portions of your proprietary information exists elsewhere. And we're just seeing it on the biotech side as well. I think it is worth noting that there are domestic CROs who can do this work.
The problem with those domestic CROs is that the economics just don't work for the volume of labor needed, right? The labor is much more expensive in the US, and so typical price points of US CROs are kind of four to fivefold higher than the offshore CROs. And so any biotech executive in the modern era has to choose: do I go with the expensive route that makes my capital allocation very difficult but secures my IP, or do I go offshore and get high-quality cheap work but at this massive IP risk?
So you're taking on this huge challenge. You've got pull in the market because that biotech down the street here in Silicon Valley would prefer going with a CRO that they trust, that can be faster, that can offer functionality and process engineering that perhaps is more even innovative to some extent compared to what they're getting from somewhat commoditized offshore CROs. How do you take on this challenge? And maybe this is a question for you, Kevin. If you take a step back, what are the bones of this business? How do you think about the physical infrastructure, the brain? How do you think about this? How do you solve this problem?
I wanted to build on one thing that Saul said. There's a quote that I thought was just super representative, from Andy Grove back in the day, warning about offshoring semiconductor manufacturing, but also, like Saul said, applies to what happened with Apple and Foxconn, and beginning to offshore low-end manufacturing, and then over time it becomes high-end manufacturing, and then we really wish we hadn't done it to the degree that we did. Andy Grove said, 'Abandoning today's commodity manufacturing can lock you out of tomorrow's emerging industry.' I just think it's very representative of where we are.
It's exactly what happened in semiconductors. You take something in commodity like memory, you start offshoring that, and today we have a situation where we have almost no semiconductor manufacturing here in the states. So you're suggesting the same thing is beginning to happen.
Oh, absolutely. The vast majority of the industry does this multiple years' worth of work in developing any small molecule drug candidate offshore, and that's just a massive amount of geopolitical leverage that we are giving away.
This is a huge problem, right? People are aware of this. It's not like they're not aware of this. So from the customer work we did, there's massive pull: they know this is a problem. They want an alternative.
But alternatives don't exist. So how do you create...
