Hi, everyone. Thank you so much for joining us for the first session of the Bernstein Healthcare Conference this year in 2026. I am thrilled to have Mike Nally, CEO of Generate Biomedicines, here with me today. Mike and I have had the pleasure of working together a little bit over the years as our careers have intertwined. But perhaps to start, Mike, for investors who are new to Generate, could you give us a little bit of a brief overview of the company, what the platform does, and where you're at with some of your key programs, and then we'll dig into those pieces a little bit further.
Cool. Well, thanks, Courtney, and it's great to be here. Generate's a company that was founded in 2018, pretty much ahead of the curve, using generative AI to design protein therapeutics. This is a company that was founded in kind of a pre-AlphaFold, a pre-ChatGPT world to use the power of machine learning to understand and program biology. From the outset, the early results have been just extraordinarily encouraging that we can fundamentally understand through big data sets how a protein sequence ultimately translates into protein function. At the heart of the company is kind of an integrated dry and wet lab-based approach. We push the frontiers of computational models, diffusion models, to create good priors of what is an extraordinarily large search space. Then we have a wet lab and a verification engine to actually build and test all of those computational suggestions.
I think if you think about the domains of generative AI where we've seen greatest promise, it's always been in cases where you have a rapid verification cycle. The reason DeepMind started in gaming was you had a scoring system that you could rapidly train models on. Why we're seeing such great progress in coding and math is a similar feature, that you have a rapid verification engine. Biology doesn't have a natural verification engine, and so having this tight coupling of dry and wet lab is what allows you to create highly performant models to ultimately drive toward a world where we can program biology.
Fantastic. Maybe kind of diving into that a little bit more, you spoke about the wet lab, dry lab, and kind of computational modeling interactions. Can you just walk us through how you use AI to create a drug, kind of choosing the target—
Totally.
—the design, the sequence, and where does AI make the biggest difference—
Yeah.
—for you guys?
Well, I think it's changing really quickly. Even since Generate was founded in 2018, there have been two fundamental revolutions in the computational space. First was the introduction of large language models back in 2022, and the second is now this agentic science revolution. In both cases, these sort of technological inflections have actually extended the reach in which we use these technologies.
Now every program that we work on starts with an agentic design phase. What we're finding is that you can basically have about 1,000 PhDs interrogate any space in biology and come up with answers in what would historically have been months in down to a week's time. Your ability to understand the foundations of biology are changing rapidly. At the same time, what also is happening is that our ability to molecularly design and come up with answers from a molecular standpoint is radically improving. I think if you were to say, where are we furthest along, it's probably in the molecular design space.
Where we're starting to understand how to, in some ways, code for molecules. I think you can see over the next 5-1 0 years, a world where molecule design will become commoditized.
Which I think has huge implications for the industry because you'll then be able to basically create the molecule for any target that you so desire to perturb. Where we're going next, I think this is partially just the maturation of the company, is thinking about how you use these tools in the clinic. We're seeing there's just a huge amount of waste in the clinical development process. Our lead asset's in a phase III trial for severe asthma. The enrollment period for that trial is two years. Nothing to do with the clinical endpoint. This is about how do you find patients and how do you find sites. We're going to 40 different countries, about 400 different sites. Each site is enrolling at a rate that's almost embarrassing.
The idea that we can't use these sort of technologies to identify patients who have clearly left digital signatures of the hospitalizations of their asthma exacerbations in a much more rapid pace.
Yeah.
I think this is low-hanging fruit for the industry. There are huge ways where we can, I think, take out the waste in clinical development. I think there's huge opportunities to radically improve the filing timeline.
Usually from study, last patient, last visit to protocol submission or BLA submission, you oftentimes are looking at a three to six-month process. The idea that an LLM can't draft your regulatory submission on the back end of a study almost instantaneously, I think we can envision those sort of worlds. There's a huge amount of waste in that cycle that could be taken out. Then where I think it becomes even more interesting is then, where do we find digital signatures and digital biomarkers to enrich populations so that the N in the study can go down and you can show greater treatment effect? I think all of these tools are going to come together, I think, at different stages of maturation. I think the complexity of biology is still going to puzzle us for some time.
Yeah.
But the reality is, I think we are on this journey toward a more programmable future of medicine.
Fantastic. You mentioned two things there that I want to spend a moment on. One, you used the word commoditized.
Yeah.
I think that's a big question in this debate around using AI for drug discovery and drug development is, where do moats lie and who's going to capture value?
Yep.
I want to ask this question in two ways. Kind of philosophically, where do you believe moats lie, and where is that commoditization going to exist?
Yeah.
But the second part is specifically for Generate. What do you think sets you apart and sets your platform—
Yeah.
—apart in terms of differentiation versus peers? That might be AI biotechs or Anthropic or some of the other—
Yeah.
—even the large multinational pharmas that are trying to play in this space as well.
Yeah. I think our fundamental premise is that we believe models will commoditize. I think where the moat lies is in data.
The tricks of a model, once they are used, can easily be mimicked, copied. We can see this in the LLM space where one day Claude is the best LLM, then the next day it's ChatGPT, the next day it's Gemini. You're seeing this kind of race, and now a number of open source models are closing those gaps as well because, in some ways, the same training data set is being utilized, the internet. In biology, I think where there's points of distinction is we only understand 5%-10% of biology, right? If you think about what are the most therapeutically relevant data sets, many of those data sets haven't been created yet.
There aren't well-curated data sets to train the models on more sophisticated parts of biology. Where we're seeing commoditization in molecular design right now is on affinity, right? You can find binders to anything. I'm not sure how important that is given the fact that you could find binders using immunization campaigns historically.
Yeah.
We were actually pretty good at that. But as you know, to make a medicine, it is a complex co-optimization equation. What you are co-optimizing on is not just binding, but it is the functional assay, which, as far as I am aware, there are no functional data sets to train models on. There are similarly a whole array of developability, manufacturability, immunogenicity data sets that are still very much ill-defined. For us, I think where you see points of divergence is how do you collect the biologically relevant data sets at scale, at pace, to better train models to come up with more distinctive answers and try and solve the molecular questions that others have struggled to solve using traditional techniques.
Fantastic. It sounds like as you speak through this, there is almost a series of low-hanging fruit opportunities with the application of AI to drug discovery. But over the long run, that kind of longevity of success will be very much determined by some of these data moats and kind of your ability to continue differentiating as well.
Yeah. I think what is exciting. Again, I think this is all a distant future, right? But if you just think about what drives biology. Proteins drive about 95% of biological function.
Yeah.
Proteins are made up of code. All of that code conveys function. So in the long term, that is a tractable interrelationship that with the help of better computers, we should be able to understand. What that, I think, suggests for the industry is over time, drug discovery migrates from an artisanal craft to an engineering endeavor.
Now, I am not sure I will see that in my lifetime. I think it will be a much longer arc until we get to pure engineerability. But what I do believe is that the ability to take steps there will modulate probabilities of success higher in the not-too-distant future. And if you can go after, as you rightly point out, Courtney, the low-hanging fruit, and take on some of the very well-known targets that have strong biological underpinnings and find molecules now using these advanced technologies that were perceived to be undruggable almost the KRAS—
Yes.
—of biologics, that has huge opportunity to improve human health, but also create great businesses out of it.
Absolutely. Maybe to spend a few minutes on some of your pipeline assets as well. You've got GB-0895, which is your most advanced asset in the clinic so far. It's entered your phase III program, which is really exciting. This is at TSLP. Can you kind of help us understand, first of all, why did you go after that target?
Yeah.
Importantly, what applications of the Generate platform at the time that you were developing-
Yeah.
—and discovering this asset, what was being applied then?
Yeah.
Perhaps, maybe even contrast that with some of the things you are doing for early assets today.
Totally. Why TSLP? The early observation we had with the Generate platform is we could manipulate the binding region of proteins in a much more profound way than any other technology on the planet. Historically, the way we have manipulated the CDRs of proteins, which are the arms that basically bind a target, is we use random mutagenesis, so error-prone PCR, or we would use computational biophysical-based techniques like Rosetta. Those could only change a small fraction of that binding region before you would find no functional variance in a library.
What we saw early in the Generate journey was that we could change up to 70% of that binding region while retaining or enhancing function. The first problem statement we asked ourselves was, is there an area of de-risk biology that we could use this sort of technology to find a better therapeutic alternative to an emerging therapy? Amgen and AstraZeneca were presenting the first data on TSLP at that point in time. We thought there was huge multi-indication potential. The one question that was outstanding in the field about computationally generated proteins was immunogenicity.
Yeah.
One of our board members, Frances Arnold, I remember her saying to me, "Mike, you don't know what is going to happen when you put these molecules into humans because nature has a way of conditioning proteins so that our immune systems doesn't reject them. Will computationally generated molecules be highly immunogenic?" That was a question nobody had ever tested before. What we wanted to do in that first case was tune down biology risk, so go after known biology so that we could isolate the technology risk of immunogenicity. What we found was we had a very well-behaved molecule.
With the TSLP story, what we were able to see is that this molecule had a very low rate of ADA and showed the therapeutic potential to be a much improved version of an anti-TSLP antibody. We were able to use the technology to drive a 20-fold improvement in binding affinity. We saw a five-fold improvement in preclinical potency. We were able to extend the half-life of the molecule to what I think is one of the longest half-lives of any antibody. It's about 98 days.
That allows you to get to every six-month dose regimen. Because of the technology, we were able to do that complex multi-parameter optimization where you're driving affinity, you're driving potency, you're driving the manufacturability and developability parameters so that you have this well-behaved molecule that has performed basically as anticipated so far in the clinic.
And in terms of that last comment that you just finished on, it's performed as anticipated in the clinic. You jump straight from a phase I to a phase III. That takes a decent amount of conviction, a decent amount of risk. You've got a massive amplification in expenditure required as you—
Yeah.
—as you take those steps forward. Can you just talk about what you had in hand as you made that jump forward?
Yeah.
Why did that make sense in this case? Is there something about this molecule, this indication, or is this something we can expect to see Generate doing in the future as well?
Yeah, I think it's case dependent, but I think what we recognized was the conditions were well set up to think differently about the clinical pathway. The first thing I'd say is we took inspiration from GSK, and I give them a lot of credit. They did this with their long-acting IL-5, depemokimab, where they basically said, "We know if we modulate some key biomarkers in asthma, they're very highly correlated to asthma exacerbations.
For the IL-5s, it was EOS. For us, we were looking to modulate not only EOS, but FeNO, IL-5, and IL-13.
If we saw comparable reductions in those biomarkers, comparable or better reductions in those biomarkers, to the marketed anti-TSLP that is dosed monthly, we knew that we would likely then correspond to a comparable level, if not better level, of exacerbations.
Part of what we did is we went into a, unlike most phase I trials, we went into mild to moderate asthmatics where the standard of care is still inhaler-based therapy. We saw, could we modulate those biomarkers in addition to generating safety and PK data. When we had the phase I data, what we saw was the molecule is very safe.
We saw that from a pharmacokinetic perspective, at the relevant doses, we were seeing sustained suppression of the biomarkers out to six months, which was really encouraging. That data now is out to a year they are still flat as an arrow. Then we saw that the modulation of those biomarkers was as good, if not better, than the marketed anti-TSLP.
With all of those data, we said, "Let's take this to the regulators." They have a pathway called the Model-Informed Drug Development pathway—
Okay.
—at the FDA. We said, with both this safety with this PK/PD data, coupled with modeling that we could show that we were basically binding the target at the same level or better than the marketed anti-TSLP, which has also proven to be very safe in the market. We said when we couple all this data together, it gives us a strong premise to go directly to phase III trials. The other thing that informed our decision was when we looked at the dose-finding work on anti-TSLP antibodies, what we saw was in the dose-finding studies, there was an 8-fold difference in dose studied and there was no difference in exacerbation rates across that 8-fold difference in dose. For us, if we were to do a phase II study, we were likely to ultimately going to make a pragmatic choice on dose anyways.
Rather than spend $50 million and take two years to do that study, we said we're better off making a pragmatic choice now and going directly to phase III.
It's kind of interesting to hear you speak through some of those decisions. You also made a decision to go from the mild moderate to the severe as well.
Totally.
Is there anything else that you would want to highlight as you talk about kind of that transition of indication?
Well, I think, again, because it's a precedented mechanism, what you know is that through modeling, you can show across a range of different eosinophils or the range of disease severities, how well do you block the target. That is what gave us the confidence to kind of expand across different populations.
Yeah.
And the fact that depemokimab was able to do that with the IL-5 mechanism as well.
Absolutely. Perhaps just to take a moment to ask about this internal ownership and internal development versus some of the partnerships, because I think this is obviously your most advanced asset. It is wholly owned.
Yes.
Compared with perhaps other assets where you have 50/50 partnerships, other places where you are doing discovery work for large cap pharma names like Novartis and Amgen.
Yeah.
Can you just talk about that decision and why was this the right asset to retain in-house?
Well, I think part of it was when you work on a frontier technology, people are always asking you for proof, like does it work? When I joined Generate, the question was we have never seen a molecule generated out of a computer enter the clinic.
As soon as you enter the clinic, then the question becomes, well, we have never seen a molecule generated computationally get approved. We were going to always be confronting those sort of questions from the cynics in this space. I think to be beholden to a partner to advance a program at the pace that you desire for your lead asset was something that was concerning to us.
We wanted to be able to answer these questions very concretely in as quick as a timeframe as possible. This molecule that is now, as you rightly point out, Courtney, in phase III, was designed five years and three months ago, and it is a year into phase III. If you think about the timeline to go from concept to phase III trial, it's about as quick as we've seen for a chronic care product in the biologic space.
For us, we're seeing the benefits of these sort of technologies to speed up the process. It was much more capital efficient. Phase III trials are expensive, and that actually leads to why we think about a multitude of different options for our pipeline, because the ability of this technology to translate into meaningful molecules is beyond what a small company like Generate can prosecute on their own. There's no doubt about it. We've put five molecules into the clinic over the last five years, and we had zero pipeline programs five years ago.
We know there's a steady stream of molecules, and that's why we did deals with the Amgens and Novartises of the world, where between them, they're helping us extend the reach of the technology, where we do the work up to lead candidate. That work can be done in a very timely and efficient way costing us $100,000 . Then we get to participate both through an upfront payment as well as up to $370 million in milestones, and then royalties on the back end of high single to low double digits. For us, balancing this where do we want to take bets ourselves? Where do we want to prove the technology?
How do we do so in an efficient way? All of these things kind of come to the fore. We're kind of now starting to have to think about confronting another challenge, like commercialization- which I know some in the room have crossed that Rubicon and certainly to me, that's a big strategic choice for the company, and one that we have to kind of step back and say, "Is Generate the best person to do that globally or not?" I think having spent a lot of my career at the commercial end of the business at Merck, big companies are pretty good at that kind of stuff, whereas I think biotechs are much better in the earlier discovery phases of design.
Absolutely. Lots of kind of big decisions ahead as you think about the potential for expansion with this agent as well. I do want to spend a few minutes on your next asset in the pipe, GB-4362.
Yeah.
This one's just entered phase I. It's a very different indication—
Yeah.
—than the first one we spoke about. Perhaps the type of therapy that we don't necessarily have in market today.
That's right.
Potentially a lot more biology risk. You mentioned that the first asset was designed five years and three months ago. This one was designed a lot more recently. Can you just help us understand what were you able to do with this asset—
Yeah.
—that you couldn't do—
Yeah.
—back then? Maybe we can spend a bit of time on kind of the potential for this asset and the types of indications—
Yeah.
—you might go after.
Some of our clinical leads at Generate spent a lot of time developing KEYTRUDA. One of the things that they were struck by was the emergence of combinations with antibody-drug conjugates were transformative for outcomes in a number of different cancer types. Urothelial cancer was one of those domains. I still remember, gosh, was it ESMO 2018?
Yeah.
Where the KEYTRUDA plus enfortumab vedotin data was presented and the survival rates were jaw-dropping.
But what was hidden in that data was that two-thirds of patients on that combination were suffering from peripheral neuropathy. Of those two-thirds of patients, 20% were discontinuing therapy, 20% were down-dosing, and about 20% were going on dose holidays. What we saw at ASCO just earlier this year was that the longer you stay on drug is directly correlated to overall survival rates. Neuropathy is the primary reason patients come off these drugs. These guys came in and said, "Well, couldn't Generate's platform distinguish and create a protein or an antibody that selectively binds to the cleaved payload of these ADCs which cause the neuropathy, but don't interrupt the intact ADC that is killing the cancer cells?" We were like, "Let's give it a try. Let's see how advanced these technologies are.
Can we distinguish that circulating toxin from an intact ADC?" What we found was we found a molecule that could do that. The beautiful thing about it, Courtney, is that this molecule, and what we've been able to show in the pre-clinical phase, is that you still get the benefits of the tumor killing without that circulating toxin poisoning nerves.
The FDA has given us Fast Track designation. The phase I study started earlier this year. It is enrolling at a really exciting rate because I was worried that if you were a first-line patient with urothelial cancer, would you want to take this sort of a therapy that could interrupt your chemotherapy? I think given the experience of clinicians who see how neuropathy is leading to permanent nerve damage in a high proportion of these patients, they thought the risk-benefit was strongly in favor of this investigational agent. We're enrolling that trial very well. We'll have, we think, the dose by the end of this year, and then what we'll do is do a 1B extension where we'll look at our ability for patients with Grade 1 neuropathy to stop the progression to Grade 2.
We think by next year we'll have proof of concept with this agent. Then from there, I think there's actually a pretty streamlined registrational path where if we were able to show sufficient reductions in neuropathy, this program could actually be approved before our anti-TSLP program.
Oh, wow.
Yeah.
That's really exciting and kind of bridges from kind of very clear pathways that have been well divined and well commercialized to perhaps a more unique space.
Yeah. I think that's key though, Courtney, because I think as we look at the application of the technology going forward, clearly there's an opportunity to optimize existing assets, and I think that can improve patient care in a meaningful way. But what really excites us is this ability to do things that you couldn't do with traditional technologies, and to go after kind of these undruggable domains and through both our own work like GB-4362 this payload neutralizer program, and through our partnerships, we're seeing an ability to modulate different biological functions in ways that you just couldn't do historically. One of the things we've seen with Amgen and Novartis is we can drive exquisite selectivity of an antibody to a target over close homologs.
That opens up this whole space of ion channels that have been really hard to drug with biologics. We can drive 10,000-fold improvements in selectivity to almost indistinguishable homologs.
We think our ability to perturb biology with precision is greatly enhanced with these sort of approaches, and that's what I think excites us most about the future of these technologies. It's not kind of coming up with a better version of an existing therapy, but actually cracking entirely new domains where patients can benefit.
Absolutely. Cracking the biology that we've just had to suffer with—
Exactly.
—to date. Perhaps just on the kind of platform kind of capabilities that you have in your hands today versus what you had back then.
Can you just talk a little bit about kind of some of the evolution of the Generate platform and what gets you most excited about what you can do now that you couldn't do back then, and what you might be able to do in a year's time?
Yeah. It's night and day. When we started the company, in some ways the models were already starting to become highly performant very quickly. Once Google published the kind of transformer architecture paper in 2017, there was a path to having highly performant diffusion-based models for biology. But the verification engine that we've historically used in traditional lab workflows was not fast enough to keep up with how many hypotheses these models could generate. Part of the innovation and, there may be as much, if not more innovation on the experimental side than on the computational side. When we started the company, you could only build and test about 200 proteins to an individual target. That is what experimental workflows would limit you to, and it would take somewhere around three months to design, to build, and test those.
Your feedback loop was running on a three-month cycle. Today, we can design as many as you want. We can build up to 10 billion, and that cycle can be down to seven days. The data velocity that we're generating today is unlike anything the industry's ever seen before. This is all through not automating existing workflows. This is through fundamental process redesign, miniaturization, microfluidics, pooled experimentation that allows you to build. These are not random libraries. These are, I have a code. Every one of those codes is purpose-built and then measured. All of that data feeds back to train the models.
For us, this experimental verification loop is moving at a pace that allows us to interrogate many different domains of biology that you just simply didn't have the data to interrogate, and it would've taken years to generate sufficient data sets. Similarly, we made a big investment back in 2022 to buy four cryo-EM microscopes, and cryo-EM is one of the most artisanal of all experimental techniques. It takes usually about three to six months to solve a structure of a protein. We now are solving a structure per scope per day. We'll have more data on protein-protein interactions in the public domain in the not-too-distant future. What's really cool is we're actually changing the type of data we're collecting now. What cryo-EM allows you to do is have a molecular microscope that allows you to see atomistic-level resolution of a protein.
We all know that proteins exist in a dynamic state, and yet all of these images to date have captured a static image of that protein. We're now able to use machine learning techniques to recapitulate the actual protein dynamics. So, you're actually now capturing movies of how the protein interacts with the target, and that's now teaching our models the statistical mechanics of proteins. Through these sorts of data sets , what you're able to do is have a much deeper and more fundamental understanding of biology than we've ever had before. So I think that's kind of where we see the frontier and with each of these sort of data sets, I always look at biology as kind of an onion. There's always more and more complexity that you're going to try and tackle.
There's greater resolution that you're going to need to see and measure biology at. But we're now developing the techniques on the most therapeutically valuable data sets to understand it and then train models on that.
Fantastic. It's a future world we're already beginning to live in, which is kind of exciting. I was going to leave this question right to the end, but I want to ask it now on the basis of what you were just talking about, is just there's a lot of noise about Anthropic and what they're doing, and perhaps they're going after drug discovery themselves. There are large pharma players making some bets in this space. There are other AI-oriented biotechs—
Yeah.
—that are investing in things as well. Can you just help us understand kind of some of the distinctive decisions, and I think you mentioned obviously the cryo, the data, et cetera. But distinctive decisions that Generate is making, as well as what you think is critical to have inside your own infrastructure to support success in kind of-
Yeah.
—potentiating this opportunity? I am just thinking about the time I toured your labs and kind of the way those labs look quite different than what you might see in other places and the type of people you have been hiring are a little different than perhaps what you would see in some of these other companies. So if you can just help give people a flavor of that-
Yeah.
—I think it would be really helpful.
Yeah. Listen, I think what's encouraging to me is that every one of the leaders of the large language model companies points to biology as the greatest impact of these technologies. I think I agree with them, right? I think if you think about what super intelligence ultimately should help us do is unravel the complexity of domains that we don't well understand today. At the same time, I think a missing part of the narrative is this question of measurement. There are huge parts of biology just that haven't been measured. Irrespective of how performant your model is, your model will struggle to understand domains that there is not data to train upon.
I think, and this is I think part of the core to your answer, Courtney, companies like Anthropic, companies like Google, companies like OpenAI, Microsoft, they have extraordinary computational talent. They are going to be great at building models with the most sophisticated techniques. Companies like Merck, Pfizer, Bristol Myers Squibb, Roche, they have great biological talent. They have built their workforces, and there's a lot of culture, there's a lot of history, there's a lot of systems that reinforce a certain way of working. The opportunity we saw as an insurgent was to marry these two worlds. It's really hard to get the best graduate out of MIT at this intersection to go to a big pharma company. It's really hard to get someone who is passionate about making life-saving medicines to be an afterthought in the mission of Anthropic.
What we promise people is they can push the foundations of computation, they can design the experimental workflows of the future to measure biology, and they have an opportunity to work on medicines that will transform the care for patients around the world. I think it's a proposition that resonates because there are a lot of mission-driven people that are in the tech world that are struggling with, am I using my life to get the next click on Facebook? For some people, that's a very fulfilling life. For many, they're looking for more. I think for us, we offer a different proposition. I think we sit in between these worlds with kind of a foot in each camp. I think the thing that's missing out of certainly the tech side is this kind of need for deep biological expertise and experimental verification.
I think that's part of what has led to, I think, some naive statements around curing cancer in five years from some of those sort of leaders.
Absolutely. To finish this conversation, value capture is really important here and as we think about the future of Generate and—
Yeah.
—where your revenue is going to come from. If you were to project 5-10 years into the future, what shape of company might Generate be?
Yep.
How much of your pipeline do you expect to wholly own? Do you expect to be generating revenue from partnerships, from the intellectual property that you've created, or from something else?
It is a question I spend an extraordinary amount of time thinking about. I personally believe that the industry is at a very interesting inflection point. Right now, just the pure fact base is large pharma pays an extraordinary premium for molecules post-proof of concept and extraordinary discounts for anything pre-proof of concept. Right now you are highly incentivized if capital is available to take things post that proof of concept readout. I think this has a lot of similarities to the tech industry back in the '80s when we saw the emergence of Intel and Microsoft for the first time. IBM had been kind of the integrated leader in technology. The integrated model was the model to pursue. If you look at all large pharma companies, they all pursue an integrated model themselves.
They have been able to do that despite the fact that their innovation engine has slowed massively because of high fragmentation in the biotech supply base. You think about the thousands of biotechs that exist, there is negotiating leverage when there is only 10 large pharma against those biotechs for products to ultimately commercialize. What this moment is changing is discovery is becoming scalable for the first time. With models and with data, you are seeing the introduction of economies of scale in drug discovery for the first time in the history of the industry. Historically, I was always trained at Merck, like mergers are bad because R&D productivity goes down, because you are disrupting an artisanal craft.
If the computer is now at the center of that process and it is a learning system, I think you probably can envision a future where you may see the value chain in some ways become disintermediated. I think that ultimately leads us to a model where we have to today, given the economic structure, take molecules ourself. In the future, there may be a very valid path where we leave distribution, which is kind of the late-stage clinical and commercial, to large pharma, but we are able to argue for fundamentally better rents because of the value you are inherently creating with these sorts of technologies .
Fantastic. Thank you so much, Mike.
Thanks, Courtney.
It has been a pleasure to have this conversation with you. I think we managed to crack some of the problems and some of the conversations and debates that are most important in this area at the moment, hopefully gave everyone a better flavor for Generate Biomedicines as well.
A lot more work to be done.