Outcomes Backwards: Jeremy David and Drew Goldstein on Palantir, Ephemeral, and Rethinking Drug Development

(upbeat music) Welcome back to the Future of Healthcare AI, I'm Devin here with my co-host, Paul. Today we're speaking with Jeremy David and Drew Goldstein, co-founders of Ephemeral, which is a new company using AI to help bioforma, discover, develop, and deliver medicines faster, and at lower cost. Before starting Ephemeral, Jeremy and Drew spent years at Palantir, where they helped the company build a healthcare business. At a time when Palantir was best known for its national security efforts. Together, they led partnerships with scaled healthcare systems like Cleveland Clinic and Tampa General. Bringing in outcomes first approach that started from a clear mission and build capabilities in AI to match it. That's right, at Ephemeral, Drew and Jeremy are applying that same outcomes-oriented mindset to rethink the drug development process. One of my favorite moments from this conversation is Jeremy's point that real organizational change starts with the willingness to tell the truth about how work actually gets done, even when it's uncomfortable. And especially when the reality looks different from the intended process. That's a theme that comes up throughout the conversation. Lasting change from AI isn't about layering AI onto existing workflows. It's instead about rethinking them from the ground up and creating a culture where people are willing to challenge assumptions about the way that things were done in the past and pursuit of the best solution. As Jeremy and Drew see it, the future will be defined by thousands of small, AI-optimized decisions that lead to better medicines, better care, and better outcomes. Let's get into it. - Jeremy Drew, really excited to have you on the program today. Thanks for joining us. - Happy to be here. - Thanks, Devin, thanks, Paul. - You guys have spent many years at Palantir really helping some of the nation's most important healthcare companies transform by leveraging AI in some of the largest health systems, as well as some governmental organizations like the Department of Defense and NHS over in the UK. And now you're taking those experiences and doing something different and something new. We'd love to talk about your new venture. But before that, maybe we can just kick off with how you got interested in healthcare in the first place. - Drew, let's get started with you. - I was in finance briefly before coming to Palantir. And the day that I joined, I got told, hey, you're gonna be flying out to Houston on Monday. Go and buy yourself a copy of this book called "Cartoon Genetics." Take that with you and you're gonna go meet with a melanoma oncologist and his research lab. And your job is to help him better understand the data that they've generated on the multiomics information about patients with UVO melanoma and help understand the implications of that for how they treat those patients. - And I didn't know a single word in that description that was provided to me. And the "Cartoon Genetics" book did not help one bit. - That sounds like a pretty intense first day. - It was, I guess the good part about it is they tell you at Palantir they throw you into the fire. I got the right first experience to set me up for that. I started on that project and very quickly caught the bug. You could tell how important the mission was to the users on the other side, oncologists and their teams. You could also tell how ill-equipped, at least at the time, folks like that were, in terms of having the tools that they needed to understand the data being generated about their patients, to be able to understand how to improve the way that they were caring for them. And I think that gap between how critical the work they were doing and how lackluster the tools were to be able to do that. That was incredibly motivating, at least for me, and as a 23 year old, showing up at Palantir back in 2018. That's great, have a good journey. I didn't get into healthcare right away, but I did have a funny first day at Palantir. I sort of indicated I wanted to travel, and they said, could you move to India on Sunday? So I moved to India for six months and then came back on one of those COVID rescue flights from Mumbai in 2020. My parents or doctors, I always sort of grew up in a household where like there was no other virtuous thing to do other than healthcare. Everything else would be lesser than, and so it was always something. I wanted to do as a bioengineering college. And then when I joined Palantir, I didn't really have a strong preference other than I wanted to go to interesting places. It was actually Drew who recruited me to the healthcare world where Drew worked with me on a mining project. Palantir does work with Rio Tinto, and we cross paths there. He recruited me to join that team, and then the seven days later, he left to go rejoin the different project. And then it was a year later when Alex had met the CEOs of the Cleveland Clinic in Tampa General that Drew got tapped for that, and then he recruited me to help him. That's great. It seems like one of the things that's made Palantir really successful is this melding of really understanding business process and what drives the businesses and then taking all these technical capabilities and harnessing AI. Can you talk about how that actually works, the methodology? It sounds like in big examples you used, you're coming in and really trying to understand how the business model, the companies, you're working with work first, then figuring out how to apply the technology, but could you just explain a little bit about how that process works? - Yeah, I always feel funny explaining it, 'cause it always seems like the simplest thing. I had a really hard time joining any company when I was at a college. It took me a while to get my way into the corporate world. And I joined Palantir basically because they explained what they do in a way that was like the first time it's made sense to me in a corporate setting, where, okay, well, if you wanna make a business better with data, AI, software, so on and so forth, you actually have to use the data software, AI in the execution workflows or in the decisions that happen every day in the organization. And then if you go to the organization and look at how it works, that tends not to be true. It's this software systems that people use every day or two rigid, so people eject out and use pen and paper and Excel and phone calls, and this is especially true in a hospital. And so it's sort of fairly simple idea where you have to go look at the world and look at the workflows and solve backwards for men, kind of reorganize them, which I think is the same thing today, and AI that's bomb-making, like economic value getting created. And then you could start asking questions of like what software infrastructure do you need or an infrastructure, do you need to actually do that well and do that quickly? And that was sort of the classic. People misunderstood what's Palantir. Services company was a software company, and then it was sort of both. - Right, I was one of those people, 'cause I was at Goldman before I'd come into Palantir, and I did not have the same immediate culture match feeling, I guess that Jeremy did, of oh, this is like the obvious way to do it. I had been taught for the 13 short months that I had survived in finance, that when Palantir touts about how they go really deep in these partnerships are huge, that sounded like customer concentration risks. - Well, that's true. Drew seems very normal to everyone, but at Palantir, he was really a freak. It was like, oh, someone who can talk to people. - It was fun to be a different kind of black sheep. I didn't really get that experience in many years, and other than the red hairs, it was a good chance to be the odd man out. So I had to learn from first principles, observing the Palantir method, and accepting the fact that it was never gonna be easy to explain our pricing model. It was never gonna be easy to explain our engagement model, but that's okay, 'cause we'll do valuable work and people will figure it out as we go. I think a lot of what you're talking about looks more obvious in hindsight, for much of the last two decades, a lot of the innovation in the healthcare technology ecosystem has been product-led. SAS software companies developing solutions, perhaps in search of a problem, but not vis-a-vis co-development with their customers, and you guys coming at it more from a, let's deeply understand the problem first, and build a bespoke solution on the back end, makes a lot of sense, and I think there's a lot of companies now that are emulating that method. So, kudos to you guys for being first and all of that. I love to go back in history maybe to when you guys started the healthcare vertical of Palantir in 2022. You guys were really the two pioneers of that. I'd love to hear a bit more about what that first year of building the business look like, and in particular, the barriers and the challenges that you had on the go-to-market side. Our experience has been that large systems historically bias to build themselves, they have very large development shops, or to consume from core vendors like APEC who are trying to orchestrate the entire system. So, I'd be curious how you convinced all of these health systems CIOs to bet on to mid 20 year olds, and a company that was better known for counter-terrorism versus going to APEC or trying to do it themselves. - The thing that I think we were very married to that is state constant, we wanted to increase the capacity of the US healthcare system, and I think there were layers to the mission that got added on over time, and saw all these things that were broken that were in the way to actually get there. And so, when we would go to whether it's a CIO or a CIO of a hospital system, we would be trying to figure out how do we get from where we are at the beginning of the conversation to an opportunity to show that we could increase the capacity of their healthcare system in some small way to start proving the value and then grow it from there. So, I think we were extremely adaptable from that perspective that served us well in the beginning, and then we very quickly got to like a real empirical opportunity to say, did it work or not? And then that was like the moment we would always win, where it was like, okay, you could sort of compare doing X and Y in the abstract in theory, but it's very hard to understand what the right decision is in theory, especially with technology, where it's like very easy to obfuscate what's going on, and then when you get it to something where it's like, okay, this was implemented and this was implemented, did it work? It became very easy to make our case. But I think is it general trend today when people think about, do I do this anthropic thing, do I do this open AI thing, do I do this other thing, and it's just impossible to reason about it in theory, if you're not an extreme technical expert in both AI and the operation that you're implementing into. It's just like how quickly can you get to something where like you get a falsifiable, did it work or not? And I think that was basically our approach. - I think that sums it up well, the part that Jeremy alluded to around making sure you're picking the right problem was something that we learned a lot over the years, but got the crash course in the first year of the partnerships where we found a way to get access to an opportunity that was on the critical path to how that health system operated. Those are the ones where we had the chance to then do the kind of meritocratic test that Jeremy described and what mattered was if we won, then we won a test on something that was actually valuable and meaningful for that organization. where it was the right starting point to continue to go deeper and partner. And then the times where it didn't work, where the times where we didn't pick the right problem, we didn't bring a meaningful enough problem. And I think that was an important lesson for us early on was if we get the chance to do this meritocratic test, if we're successful, what does that imply for the business? And if it implies something meaningful, then we're on the right path. If it doesn't, then the test doesn't really make sense to either party. And I think that was a really important lesson for us in the early days. Fascinating. I'd love for you guys also to maybe unpack for our listeners what your sales process for pitching a pilot and then the tiger team deployment of executing a pilot or a new product build looks like inside of health systems, what that 90 days looks like. For having work with you guys inside of some of our portfolio companies, it is truly a very distinctive approach where you guys get in a room with a small subset of individuals, you code overnight and seemingly pull these rabbits out of a hat where you have a miraculous pilot and a 48 hour period so people can see a tangible proof point and then follow through to execution at speeds that people have really never seen. But it'd be great if maybe you guys could talk a little about the model for how you execute to value so quickly why that's different than other solutions. I think there's some X that's the end value to the customer. And then there's Y, which is that customer's perspective on what they need to do next. And oftentimes those are misaligned in different ways and different parts of the organization. Maybe there's an organization within IT that's main goals to put all the data and data bricks or snowflake. And it's like, okay, well that's sort of their Y. And then maybe there's an operational person who's trying to solve for a very local problem that might not move the needle. So you have to hold two things in your head. There's one where it's like what's going to meet these folks where they are and meet the thing that they're trying to accomplish. And then marrying that with our working opinion and then oftentimes the alignment with the CEO of the company of what's the thing that actually moves the needle for the business. So I think that's what sort of defines what we try and work backwards from. And I think our focus there is helpful because if your goal is to integrate all the data into a data lake or something more abstract that's less aligned to a specific value outcome could work on that for six months and not really get anywhere. And I think there's tons of examples with AI. It's like, I've rolled out my cloud co-work to my entire enterprise and it still seems to work the same way. That's sort of the modern day equivalent of that. Whereas if it's like, oh, this like the civic workflow is now generating way more output with the same input. There's something to solve in 48 hours rather than it being this very nebulous thing. And then there's certain cultural norms around tenacity and grit that I think are helpful. And then there's a product piece of it too. As I was saying earlier, there's some X that is a product you need that marries AI data software into these complex intricate enterprise operational workflows. And Conter did a good job at making that product and then there are different products. You might use different situations, but you need leverage to be able to do that. Humans alone can't do it in 48 hours. It's like the humans, the culture, the trade craft and then the leverage that comes from the product. Palantira has really expanded the business with a lot of the top health systems, the customers that Palantira has in healthcare system space accounts for 16% of US hospital admissions. So it's pretty substantial sample set of the US population. I'm just curious if you're able to see patterns across customers that you can then help apply to individual customers. Obviously, these institutions are really big in their own right. But we've seen across some of our healthcare IT businesses when you have got a network of facts across different types of providers and different geographies. You can really drive a lot of insight, a lot of impact on efficiency of healthcare, quality of care. Drew, at Palantira, we're also looking across institutions and geographies to identify common patterns and challenges. If so, how did you put those insights to work? - There are different levels of granularity to look at this from. And I think there's one level of granularity that's, oh, is it revenue cycle or patient flow or nurse staffing or clinical research or more kind of the buckets of problems? Every health system is at some level of abstraction working out a lot of the same problems. I think the more interesting granularity is the underlying specific challenges that lead to what makes it hard to deliver on the vision or the intended outcomes across those different areas. We would see this by actually observing an operator and watching the person that's responsible for managing appeals in the revenue cycle space or the person who's in the transfer center trying to place a patient that needs care within the hospital. If you watch the way that they do their work, as Jeremy alluded to before, you see a lot of folks that are kind of ejecting out of the source system where somebody that's far away from the problem believes those decisions are being made. But in reality, the decisions are being made across spreadsheets and emails and phone calls and teams messages and all the different data points that have to be collated with each decision. I think one challenge was always, how do you bring that information together working backwards from what's needed at the point of the decision? Then I think another challenge was oftentimes a lot of workflows in the healthcare added up to a set of rules that a user has that are often been their heads. There's a lot of intuition. There's a lot of kind of key person risk in healthcare from folks that have been in that role for 25 years and if they go out of the office for a week, the whole place scrambles and almost shuts down because they don't have that person who's been doing it forever. And so a lot of the work that we did was how do you actually codify those processes? How do you build into a system? What those rules are? So that even if you're able to integrate all that information, you now have the ability to start to layer on the automations and the workflows where you've now collected all of the business logic that's needed to be able to make those decisions. And so a lot of it amounted to the expertise, the process rules bringing that together with the information so that you had all the context in one place when it came time to make a decision. Super fascinating. I'd love to take a step back and maybe touch on the impact that AI will have on clinicians. Clinicians are obviously incredibly important stakeholder. They're at the center of healthcare. The most supply constrained input and a key driver, not only the productivity, but also the cost of the system. I know you've spent a lot of time observing clinicians doing their work. I'd be curious, Jeremy. What do you see as the top use cases for clinicians to utilize AI either to augment their capability or to substitute and reduce the unit cost of care? I'd also be curious what you think that means over the next couple of years for how we should be training the next generation of physicians to be AI capable. So maybe to start with how I think it's being consumed today. And I think most of this is basically true if you're talking about a physician or a scientist or any sort of highly technical user, like technical in their field. So a physician today primarily uses AI as like a very effective search tool where you would previously have to get information from your database, maybe the EHRs, the most common in the health system context. And then you'd have to look at precedent in literature and things that are understood about this clinical pathway. And then for a scientist, it's things that are understood about this biological pathway. You're trying to make a drug out of. And the folks who were great at it were at extremely effective research workflows had an extraordinary recall ability from years of experience. And then there's sort of like an ethereal taste or judgment or intuition around it. But I think the way AI's made a very big impact so far is the search workflow is way more efficient. You sort of get the answer with everything cited more or less. And I think it hasn't gone much past that, honestly. Like I think that's sort of in the clinician workflow. There's a lot of good stuff about automating note-taking, I guess that is certainly helpful. It's been a lot of automating these little things like research or transcription that do matter, but perhaps won't change the overall cost of care or quality curves of the country. There's a lot of things like this in AI where you roll something out and then your employees go home at four instead of five, but it hasn't really changed the some total of your enterprise, which is great. You know, you love your early Fridays, but from an economic value perspective, it maybe hasn't moved the needle. And then I think there's like a different way to look at how AI works in a patient journey, but you could draw this out for any enterprise. There's like a set of bespoke decisions or execution workflows that happen throughout a patient's journey that involves clinicians, nurses, other stakeholders in a hospital. And those can be codified, but are very hard to codify in like traditional rigid software because there's too many different pathways. There's too much judgment. It's too hard to figure out what the computer can do versus what the human can do. I think the bigger opportunity for health systems, this is for any enterprise, is how do you actually create the execution workflows that span many clinicians, nurses, the patient themselves and other stakeholders so that you can define what the computer can do, what the human can do, what a traditional model can do. And then I think if you do that, the execution environment itself has changed. And AI is now pouring more gasoline on that fire versus being something more on the side that's a time savings type of thing. And then on the training question, I think in a lot of fields, your absolute recall is less. This classical, like can you get the perfect score on the test is probably a little less important? You need people who are able to see things in a different way and impose that taste and intuition and drive the system where all the context and all the information is brought to you. And most of the work is happening to the background and how you organize the world is the most important thing. - They're just raising an interesting point there, Jeremy. The implications of being able to implement a gentick AI across workflows, across departments, across functions within health systems is really promising, but it's challenging because you have those different departments, you have regulations, you have short ways of clinicians others working in the health system. What do you think is the biggest barrier to that vision that you were just articulating? I think there's a cultural human one. You just need folks who are willing to cut through maybe trees too friendly for that one. You have to not be afraid to offend the patient flow manager. You have to always want to tell the truth. And then there's a technical thing. A very smart fellow on our team named Mark Bessel said this the other week that I think is very clever. Why have coding agents done such a good job for us? It's like a very simple execution workflow in a single player game where your execution environment is your file system with some API calls out to GitHub or whatever. It's fairly simple. It's contained. All the information can exist in one place and the workflow is very easy to define. And it's like I said, single player. And then in an enterprise is way more complicated than that. There's a multiplayer game with a much more complicated execution environment that spans lots of source systems and files and conversations and decisions that are hard to model in one computer. And so I think there's a technical product challenge that's like, what is your execution environment in an enterprise? There's a lot of folks who are implicitly making this decision versus explicitly making this decision. Where it's like, my execution environment is like my SharePoint that my cloud co-work interacts with. That's not a very good execution environment. That probably won't work. But it's like, okay, well, then what is the right answer there? I think there's a new enterprise AI architecture question you have to ask of like, what is your answer to that question that most folks don't answer well? Just to add on the cultural part, I think part of that is on us, right or on Palantir, on the service provider that's coming in and trying to solve the problem. But there's a huge part of it that's on the other side. We were fortunate that the first two partners we had in the health system work at Palantir, Tampa General Hospital, and Cleveland Clinic. I mean, John Corris and Tom Olovich took a shot at working with as you guys put it before, like this counterintelligence company that hadn't really done anything in health care before in a couple of 23-year-olds that they shipped out to go and sit side-by-side with their operators. They had the confidence and the belief to say, hey, we're going to try something different this time. And it was their word that gave us the opportunity in those early days to actually try it. I think now with the business that we all build at Palantir and with the teams now carrying forward with the health systems, it's maybe a little bit less of going out on a limb to say, hey, we're going to try to go do this with Palantir because there are all these great case studies and stories of what's been done to date. But you do need the kind of heretical leader that's willing to do something differently early on. And I think that was a really key ingredient to get around some of those barriers that might exist in health systems. You guys recently made the decision to start a new firm, a femoral focused on the pharma ecosystem to bring some of the same automation capabilities that you guys have been talking about to really complex pharma value chain would love to hear a little bit more about your vision for the company and where you hope to take it. In many ways, the company was very underscoped when we started. I think the original idea was something like it seems like medicines are slow to make and expensive to make. We had some initial conversations with folks who are in charge of making medicines and got into some of the specifics of why it was slow and how it could be made faster. And then we wanted to start something where the mission was very clear where we want to help make medicines happen much faster and for much less cost. We really didn't have too much more than that. We've obviously learned a lot and have a bunch of sort of wonderful early partners that have almost been our version of the two brave leaders to take a shot on something that's newer and sort of take a big swing to change the unit economics and pace of how medicines get made. I think the themes are very similar to health systems in that the primary way folks use AI is sort of on the side of the workflow. You've done all the work and you're automating the creation of the slide deck would be the caricature of it. And it just turns out that that's not the important piece. The important piece is the complex set of interconnected decisions that all sum up to making medicines faster. And I mean, in different parts of the spectrum, I think you feel it in different ways. In the discovery side of the world, a lot of the existing providers have tried to reduce down the discovery process into one API call. And the discovery process is much harder than that. There's a lot of interconnected workflows. Yes, the scientific foundation matters a ton and that's usually driven by the intuition of some very smart scientists versus a large language model. And then there are some predictions along the way that can be helpful. But in some ways, it's more a factory than it is like a single idea that gets generated. Where you're stringing together experiments, hypotheses, you take the results, you come up with a new set of hypotheses, you designed an excited experiments. And then after this really hard scientific problem, you have this really hard operational problem, where you have to run a trial, which is basically operating a part of 100 hospitals instead of just one hospital that you just met last week, which is very challenging. And I think for that, there's definitely not one magic decision that makes it go way faster. It's like thousands and thousands of smaller decisions that get made on the specific participant and specific site engagement lead and specific CRA level that sum up to is your trial happening quickly to a good quality, so on and so forth. And then there's a ton of challenges on the commercial side as well. I think it's similar to how we define the health system business of balance errors outcomes backwards. We have a clear mission and we evolve what the capability set is to go and accomplish that mission with each new encounter that we have. I think the nature of being our own company, fully foreign building our own product from scratch. The rate of learning has just been incredible over four and a half months. As Jeremy Mentry has had partners that have been willing to give us without any background necessarily in the space, very deep access to sitting side by side with the folks that are moving the process along across the discovery efforts and the trials and getting the chance to see here's an opportunity to make this move faster, but show up tomorrow with something that we could build to see if it actually moves the needle. And usually it's maybe 40% right tomorrow, but we get more feedback and by the end of the week, it's actually making meaningful difference for them. I think that kind of day-over-day iterative cycle learning that's a man that we've definitely taken with us to the life sciences world. That's very exciting. I know it's early days, but I'm just curious what your hypotheses are around the areas that are going to be most fruitful for leveraging AI across this very complex value chain. What are you most excited about based on what you've seen so far? I think there's a ton of value to be had across the discovery workflows. Maybe the frame there is. It's really hard to get to an IND. It's really hard to get from a vague biological concept or an idea for a set of screens that make up your discovery platform to a true product concept that's ready to go into human for the first time and you've done the preclinical work. And I think you can certainly picture a world where you have a system, like a complicated intricate industrial system that could generate those much faster. Clinical trials are always going to be people heavy, but it's similar to running a hospital better with AI. There's a lot of leverage you can get by codifying that operation and imposing the correct decision structure at each step and being more proactive of taking the next step. The quality of the decisions matter a ton. What sites are you selecting? How are you designing your protocol? There's all of those areas that I think matter a lot. There's a lot of conversation on do we need to be doing Cuban trials or what's AI going to do to that? I don't know. I tend to be slightly more pessimistic and at least the short to medium term on that. I don't think there's been a lot of progress there, but you never know. Well, you guys step back and think about drug development in the future, say five to ten years down the road. What do you think looks different in terms of how companies are organized to tackle these complex problems? How do big far ever companies look different? How do biotech companies look different? How do more tech-oriented AI discovery companies look different? How does this all evolve from your perspective? Some of the things that stood out to us in the early days, I guess we're still in the early days, one thing that was hard to believe was the value of each day, the value of each month when you're running one of these clinical trials. The amount that people pay to get four months off in a priority review voucher from the FDA shows how important and valuable each day and each month is, and yet so much of the operation is being outsourced by these biopharma companies to CROs that charge in a way in which they make more money if it takes longer. We saw a lot of things that were maybe broken about the incentives in the health system world, and I think this is one that seems like there's opportunity to be addressed, and I think a lot of biopharma companies are starting to ask the question of, "Should an AI allow me to do this in a different way?" There are different ways in which that could manifest over time. You might have some CROs that are going to see that opportunity themselves and say, "Hey, we're going to do this differently and maybe take more risk and go more outcomes oriented, and those would be CROs that we'd love to partner with and be a part of that movement." I think on the biopharma side, you're going to have folks that are going to maybe try to take more of this in-house and be able to do that work. To Jeremy's point, you'll still need people, but I think you might be able to now do more trials with the same team and take on more opportunities as a business. One thing we've seen is it's much easier for us as an ex-pallanteur building an AI business to go and raise money. It's not easy for biotechs that have incredible ideas that could change the lives of patients to go and raise money. That funding environment is just very different. If we could find a way to cut the cost of getting to bringing a medicine all the way through clinical trials, there should be a world where you have a lot more great ideas that are getting funding because you don't need as much funding to be able to go down that journey. I think that's one thing that we'd have an ambition. Can we help change the underlying economics of what it takes to actually bring a medicine to market and does that encourage more folks to go down the path? It's a really inspiring vision and hopefully an outcome that we can all help to drive toward. Maybe the close would love to get your advice for entrepreneurs that want to step into the healthcare AI range. and in particular any advice that you'd have for aspiring entrepreneurs on how to get started, how to find product market that, and if you want, you can make a recruiting pitch for a femoral to Jeremy. Let's start with you. We've been entrepreneurs for all the four and a half months, so I'm not sure we're fully qualified, but I guess some things that we think about is be valuable in the specific first. There's a lot of pressure to answer questions like how does this scale and to label yourself very early of like what thing you're doing or what type of company you are like, which thing do you fit in? And I think that can be very counterproductive to the actual thing you need to do in the early days, which is trying to be valuable to someone. It's not to say you have to do that for the rest of your life, but if the things you'll learn through actually winning will be much more important than getting the perfect plan upfront, you have to do the company you know how to do. I don't know that I could tell someone how to get product market fit, so I don't know if I know how to get product market fit. Someone's asking advice, how do you get people to get really bought into the mission or the culture like work really hard? And I was like, I have no idea, but we just go to work every day and actually care about the thing. Some companies people want to work out because they're technically elite, some because there's super mission driven. You gotta figure out what your thing is through what are your wise words. You said winning, I think that was something we learned and volunteers, you just got to find a group of people that you think you can win with on a mission where it matters if you win or whatever that is and however you define that to you. And if I think of you surrounding yourself at the right people, you're gonna learn from them, contribute to them. It's not always the right people in the sense of like who makes you happy. I don't think I always make Jeremy happy. He sometimes makes me happy, but not always. - We haven't hugs in six years. (laughing) One day we'll get there. I think it's surrounding yourself with people that push you and have spikes in the places where you have weaknesses and where you can learn and also have the opportunity to lean into the things that you're really good at. I think that sometimes underappreciated when you're thinking about what adventure do you wanna go on with what you know, crew of people to go do it. But I think that's something that we really look for a lot when we're thinking about how we group people together, how we pair people, how we think about what groups of folks will work best together. Having the agency to think that way when you go look at, all right, who's my co-founder gonna be? What company am I gonna go work with? What projects am I gonna go work on? I think that's a really important ingredient that sometimes folks miss. - Where do you see the biggest gap today between the promise of AI and Biopharma? And how it's actually being used today in practice. The rhetoric to reality gap in the Biopharma world is extremely wide where it's like you'll see a huge press release of billions of dollars of commitment to AI partners and then what it actually means if you say it in truth, the translation for the headline, it's very minimal of how it's used. It's more like I do my literature of you faster and I make my PowerPoint slide faster and go home it for 30 instead of 530, then like we're having this lump sum engine that creates biological discovery significantly faster. I think the single way to describe the company is we wanna close that gap and I think the only way to do that is very simply to go understand the specifics of how it's done and every day make it happen a little faster and then prove yourself and make happen even faster and then go from there. - Really incredible. We're inspired by your vision and wish you an affordable luck as you push the boundaries of what's possible for drug discovery. Thank you for being here, really appreciate it. Well, that's where we'll leave it for today. As always, we'll keep exploring what real innovation looks like in AI in practice, not just in theory. Thanks so much for listening.