Weave’s Brandon Rice on Rebuilding Drug Regulation with AI NEJM AI Grand Rounds · 15 Jul 2026 Source: https://ai-podcast.nejm.org/e/weave-s-brandon-rice-on-rebuilding-drug-regulation-with-ai/ Transcript: https://podcasts.opennash.com/shows/nejm-ai-grand-rounds/weaves-brandon-rice-on-rebuilding-drug-regulation-with-ai-222467/ (via OpenNash Podcasts, machine-generated) An IND, if you go and outsource that, and this is after all the experiments have already been done, that's going to cost you about half a million bucks just to produce the package of documentation. It's going to take you three to six months. An NDA, that market application on the other side, you're talking millions of dollars and a year or more to pull that information together. whether it's an order of magnitude or two bigger than an IND. So those are the stakes when it comes to just the paperwork part of this. But there's another way to think about it, which is why we picked this problem, why I'm excited about it. Regulatory drug development is regulatory development. They're synonymous. And so if you want to think about what in drug development is pertinent to the regulatory process, it's literally everything. and so we're starting here at the goal the destination is you want to get your ind done you want to get your nda done and then working backwards into what should you do and that is something that we'll be able to provide unique insight on and as our customers build more and more of their submissions within our platform they'll have better insight into what works and what doesn't and we will be able to provide better insight to the industry as a whole about what works and what doesn't, and thereby make everything more efficient and faster. Hi, and welcome to another episode of NEJM AI Grand Rounds. I'm your host, Raj Manrai, and I'm here with my co-host, Andy Beam. And today we are very excited to bring you our conversation with Brandon Rice. Brandon is the CEO of Weave Bio, and his company is using AI to help streamline drug regulatory applications. Andy, I thought this was a really interesting conversation. Brandon had some, I think, really measured answers where he's drawing back on a few decades of technological progress. And it's clear he's very thoughtful and that he's not just recklessly throwing AI at a complex problem or at this sort of bureaucratic complexity that is very present. And so he was very thoughtful, really, I think, had some great answers And this is a lot of fun to just pick his brain about how he's approaching this with Weave. Yeah, Brandon was great. I think that what has struck me about the past few conversations that we've had, this is meant to be the compliment that it is. But there are like so many important but kind of boring problems in healthcare that actually could unblock huge amounts of value and huge amounts of benefits for patients. And so Brandon's company is working on how do I submit things to the FDA to get my clinical trials going? Brandon walks us through why that's an important problem, how hard it is and how time consuming it is. I tend to be in the reckless AI category as a person myself, where I'm excited by the big shiny object. But I thought Brandon did like a masterful job at explaining like, hey, let's take all the hype out of the conversation. Here's a really important problem. We're going to be laser focused on that. And we're actually going to get drugs to market faster. And so he did a great job explaining that. And you can tell that he's an OBS kind of person. And when he sets down to solve a problem, he really sort of goes at it 110%. So really enjoyed the conversation. And I myself learned quite a bit about the drug approval process. The NEJM AI Grand Rounds podcast is brought to you by Microsoft, Viz.ai, Lyric, and Elevance Health. We thank them for their support. And with that, we bring you our conversation with Brandon Rice on AI Grand Rounds. Well, Brandon, thanks for coming on AI Grand Rounds. We're excited to have you here today. My pleasure. Thanks for having me. Brandon, great to have you on the podcast. So this is a question that we always get started with. Could you tell us about the training procedure for your own neural network? How did you get interested in AI and what data and experiences led you to where you are today? Yeah, I'm happy to talk about that. I will try to do the very condensed version because it's kind of like the whole arc of my life, I think, that makes it make sense, but I'll do it very quickly. I was super interested in computers and science from a very young age, tinkering around with them, building my own computers, learning a program when I was a kid, and then went to school, studied biology, got out, had to figure out what I wanted to do with my life and didn't know initially. So I joined a national genome sequencing center, Didn't know anything about sequencing when I did that and did an internship there. Got a job there after that as a software engineer. And that got me onto the track to want to study genomics and bioinformatics. And so I went from there to grad school to pursue a PhD in bioinformatics. And I was studying Neanderthal inheritance in modern humans, which was super cool. but as it turned out a little bit abstract for my taste I wanted to do something a little bit more tangible and immediate and so like a good entrepreneur I dropped out of my program went and started a company right out of grad school which was a genomic services company doing de novo genome assembly as a service and that was how I got going in industry went from there to a diagnostic company to a drug target discovery company to a drug manufacturing company. It was by virtue of those experiences that we formed a thesis for Weave, which is what I'm working on now. The way AI came into it was I was leaving my last company and looking to start something new, met my now co-founder, Ari Caroline. He was the one who got me wise to LLMs. I knew the problems that I wanted to solve, the business I wanted to work on. When I met him, he was like, there's this thing, it's LLMs, it's going to change everything. And I was like, what's that? What year was that? What year was that? Late 2022. Yeah. And so he was already wise to all that. He'd been doing a bunch of NLP stuff. Earlier in his career, he was the chief analytics officer at MSK for 15 years. And so he had already been dabbling for some time and understood the technology and the potential that it harbored. And so he was the one who told me about it. I said, that sounds amazing. Let's figure out what we can do with that. We got together, started Weave, and we've been building since. Amazing. So we're going to dig into Weave a lot. But maybe before we go there, can you take us to that, a little more detail about that decision to leave grad school, leave your PhD program, and then I think you said you You founded a new company, right? Yeah. So can you maybe talk us through, because there's a lot of folks who are also finishing up med school, grad school, wondering about next steps. I think the era now is very different than it was even five, 10 years ago with LLMs and with tools that we can now use to bootstrap new ideas. But maybe just talk us through your mindset and your approach to starting a company versus joining, let's say, one of the big genomics companies at the time after you left grad school. Well, it was at least a little bit of serendipity in my case, because there's some technology that we had been working on in the lab, which was a kind of special method of library preparation for sequencing that had some potential for a lot of different applications. And so we had that technology was right in front of us. There was an angel investor who just kind of like came into our orbit at that time and was trying to find exciting science companies to invest in. So it was just a perfect storm of factors. And my PI and I decided, let's go and see if we can make a business out of this. And I obviously had no idea what I was doing. I was a grad student and my PI was super supportive. of he stayed at the university and he was like, look, if it doesn't work out, you can come back and finish. And I said, great. That was instrumental in my making the decision to take that plunge. But I didn't end up looking back afterwards. And, you know, it's just a bunch of figuring it out one step at a time, solving problems, ultimately ended up being super fun and got me onto the track that I'm on, which I find very gratifying. Did grad school prepare you at all for starting your own company or not so much? In a certain way, I would say yes. And that way is I showed up the same way to the company as I did initially to grad school, which is I don't know what the hell I'm doing here. And I just have to figure this all out. And it seems like there's a lot of much smarter people than me all around me. And everyone seems to know what they're doing. I now know that nobody knows what they're doing and everyone's just figuring it out as they go. And it was the same thing with the company. You just showed up and it's like, well, how do you do this? And you got to start talking to people, trying things and figuring it out. And so in that way, it's very similar. You're just showing up to something you have no way to be prepared for. And you just got to dive in and start going. Awesome. So we want to dig into your work at Weave now. And so started the company out of grad school, worked at a few different companies over the years, and now you are the CEO of Weave. So maybe first you can tell us a little bit more about Weave, what Weave's mission is, why did you start the company? So what we're trying to do at Weave is modernize the regulatory process in drug development. That's the most succinct way that I can put it. And in doing that, trying to get drugs to patients much faster and much cheaper. In drug development, regulatory is, it's different from what it is in other industries. In many industries, it's a fence around what you're allowed to do. And if you step outside that fence, you get fined. So you just stay inside the fence. Here, it is more like a railroad. The regulatory process is rails, and that's the path on which you get the drug to market. And there's a bunch of important milestones along the way. So it's fundamentally different in that way. It's the heart or the backbone of drug development is getting through this process. That was what a big part of the appeal was. And that was something that I noticed in the course of my experiences at those other companies, which again was diagnostic company, a drug target discovery company, a drug manufacturing company. There's this common denominator that was somehow, I noticed, extremely important. And it's the thing that everyone wants to put on their resume. I did an IND, I did an NDA, and yet it's totally overlooked and neglected and in some ways loathed when you have to go and write up all this paperwork. And so it's that sort of duality that was really interesting to me and looked like an opportunity. It's something that every drug has to go through to get to market. It was, in many places, still is an extremely manual process. And it happened to be the perfect match for the technology, the LLMs, because it's a bunch of transforming information from one form into another, building a narrative in that way. And that was just, just looked like the perfect match. Brandon, could I hop in here? So in a previous life, I helped start a drug discovery company. And the first time I heard IND as like a drug development lay person, I had no idea like what that was. Could you one tell us what that stands for and why that is such a significant milestone for a company developing a new drug? Absolutely. So IND is Investigational New Drug Application. And in the United States, that is the submission that you have to send to FDA explaining your drug to them effectively and what studies you've done so far, how do you manufacture it, what disease indication is intended to treat. And then FDA decides if you're allowed to go to clinical trials or not. So that's kind of the gateway to clinical trials effectively. that's kind of the beginning of the proper regulatory process the end is getting approval for the drug that typically happens 10 to 15 years later that application in many cases is called an NDA a new drug application so you got these three letter acronyms bookending this process which is very tidy and gratifying in that way and it's the starting line and it's the finish line and then there's a bunch of submissions along the way where you're communicating what's happening with your trials to FDA. And it's the same exact thing in Europe, in Japan, in China, with slightly different labels. Could you give us a sense of what's in an IND? Like what actually is in this packet of information that you send to FDA? Yeah. For an IND, it mostly boils down to two things, typically. one is all of your non-clinical studies. So all of the animal studies and vitro studies that you've done to understand pharmacology, pharmacokinetics, toxicology, you're pulling that all together, distilling it down to as concisely as you can into a narrative that goes to an FDA reviewer who has to read that, understand it, and see if it's compelling, if you have a compelling case to test your drug in humans. So that's the one major part is all of those non-clinical studies. The other major part is how manufacturing information, how are you gonna make the drug? How are you gonna test it? How do you know that the second and third and fourth time you make it, it's the same as the first time and having consistency and measurability there? Got it. And so this is actually a gating process. Can you fail the test and therefore you do not pass go, you do not go to a phase one clinical trials? Is that like kind of the right way? That's, that's exactly the way to think about it. So it's critically important and getting a drug to and through IND is a major milestone for any drug and, and any young drug company. What's the average success rate for an IND? over 90% is is what it is and that's because people put so much care and effort into putting these packages together you don't go to submit your ind generally until you're ready got it cool so could you tell us a little bit more about the kind of existing pain points with these submissions and like where we fits in and what you're really trying to solve there's three areas of the problem all of which are related that we're trying to solve for one is fragmentation of the knowledge and i use that word knowledge very intentionally not data but after data has been analyzed and interpreted and turned into something that is now understood about the drug. That's what I'm calling knowledge. And that lives all over the place as it relates to a drug. There's some in an academic lab, some in the company that was formed, some in a CRO that's doing studies for you. And just getting all that information together to produce these packages, that's the first thing you got to do. And that's a bunch of chasing files and talking to people and figuring out what has been done, where's that information. That's one part. The The second part is then composing that into a narrative is the simplest way to think about it. And that's what these submissions ultimately are. You're telling a story to FDA or to whoever the health authority may be. And that is a lot of summarizing and writing and tabulating information to help FDA understand in succinctly a way as possible what was done and what it means. It sounds like an area where large language models are quite useful. This is exactly why it seemed like such a perfect match for the technology. And then the third one is the communication process itself. Once you've built that narrative, the way it works today is you package all that up into a bunch of PDFs, and then you send that over to FDA or to the health authority. Those are the three aspects of the problem. All of those, in my view, are addressable with the technology that we have in LLMs and all of their kind of derivatives, like agents and things like that. And with a software platform that pulls that all together and creates a common space where people can come together to work on these things. Can I ask a question? Often in drug development, people are like, oh, how are you going to speed up clinical trials? It's great that you can get to clinics faster, but you're still like slowed down by the regulatory process. So Weave is obviously addressing that. Do you think FDA on the other side that one day there could be an LLM reading an IND application written? Like, is it going to be essentially LLMs talking? How do we know that that's not happening today? Well, I mean, that's kind of where I was going. I was going to invite him to speculate versus claim. Okay, okay, okay. Yeah. It is happening today. It's happening already on some levels. Today's FDA has an appetite for AI and is very interested in figuring out how to apply these technologies to their problems, which is sort of the opposite problem, right? You get this big bolus of information, you got to pick certain pieces out of it so you can understand what's going on. So FDA already has their own initiatives going on around this. And my hope and kind of like the distant future that I see, hopefully for Weave, but certainly for the industry in general, is that the underlying nature of this interaction can be fundamentally different. It doesn't have to be chucking PDFs over a wall at each other. It could be something that is mediated by LLMs and is a much more dynamic and interactive process. is that back and forth between the drug companies and the health authorities is what a lot of the time is. You're waiting for questions, they're waiting for answers. And so if you can speed all of that up, we're talking about months or years off the timeline. Yeah, it's a good point. Often we think of the regulation process as like semi-adversarial, but my experience talking to folks at FDA is they actually want new drugs to get to market and there is a willingness to be helpful. And so maybe AI mediated dialogues between drug companies and the FDA would expedite a lot of these challenges where instead of just throwing a 200 page PDF packet, like you said, it's much more conversant than what we're able to do today. That's right. And it's, and just to add a little color here, because you said 200 pages, the anecdote that people love to talk about here is it used to be that you would fill the back of a semi-truck with all of the paperwork for your NDA, that market approval application I mentioned, and you would literally ship it to FDA. So it was a semi-truck full of paper. And people talk about that now and they laugh and it's hilarious. But what's ironic is it's all now transmitted electronically, but it's the same quantity of paperwork. It's just PDFs now. And so the first thing that happened is we transmitted it electronically. That was a huge step forward, but it's still the same quantity of paperwork. What's happening now, the opportunity now is to reduce that to only the information that you really need. And that's what we're trying to enable here. Can I ask you, Brandon, to talk about the way you think about the core technology of the company? Because it's just amazing, honestly, to be a researcher in this time, to be developing new methods, new analytical approaches, because the tooling is just growing so fast that every day, maybe every week, actually, it feels like more than that at a higher frequency now, there's something new or a new way to think about using agentic AI tools to get what you need to get done. And if you look at the last few years, I think it has to instill in me a sense of humility. It's very hard for me to predict that the pace of useful tools to do things in my life as a professor would be at this state that it's at now, where I'm using these tools all the time in basically every part of my job. and things that I built a year ago that I thought were defensible and would be things that would last a very long amount of time or just sort of obliterated in the next incarnation of either the frontier model or the tooling that the frontier lab is building around it. And so Anthropic, for example, has really, I think, changed a lot of folks' minds around the usefulness of these tools with Cowork and Cloud Code and tools like that. And so I guess my question for you is that it's both very exciting and it's also very dizzying because you can make big investments in sort of the previous incarnation of the technology only to get replaced in a sense with sort of the next incarnation of the model. How do you think about that tension in the way you approach building technology for Weave? What you would have done for the LLM or the custom LLM pipeline in 2023, probably very different than 2024, probably very different than 2025. And then now with our current tools, where is the durable advantage or the durable moat, if you will, that you have for protecting Weave and for maintaining the work that you're doing for the next few years? So this is something that we had to learn lessons around early on. and we made some missteps that helped us learn those lessons. Very quickly, I would say within the first six months of building, the place that we got to is build with headroom and build around the models, I guess is the simplest way I could say it, which seems obvious in hindsight, but we just expect the models to get better and eventually to get really, really good. we developed the discipline to stop ourselves from trying to build software that mitigates the weaknesses of the models of that day and just expect that they're going to get better and better and better and imagine the future where they're amazing and can do a lot and what's still left to do at that point in terms of software what is our role then in that future and that's been serving us well. We just create a clear lane for the models to get better, and they have been continually getting better, and so that's been working great. We think about the models as an engine. We're building the car. You can take this thing and it gets more and more powerful over time. You can go faster and faster. You still need to steer it, have air conditioning, keep the rain off of your head, have a place for all the other passengers to sit. And so that is the platform that we're building, is all these other pieces that create a space for collaboration, which is kind of the central nature of the process we're trying to enable. And then let the LMs or agents or whatever the future instantiations may be come in and work alongside humans in that. And whatever they can do well, we let them do it. Whatever they can't do well, we let the humans do it. and we build this so that it's easy to flex between those two modes as the models continue to get better. So I think if I'm to try to parse that or sort of condense it, you're trying to build an ecosystem that even as the core technology gets better, your ecosystem will itself get better and continue to have value almost independent of the core reasoning abilities or the core capabilities of the frontier models. And I think that makes sense. I think that's the philosophy, I think of several companies that are embracing this or building on top of LLMs. I guess maybe another perspective that I have on this, and I wonder what your thoughts are. Do you think the moat here, in addition to that tooling and that ecosystem, do you think it's trust or the relationships that you have with the CROs or the pharma companies? How much of this is that they just know who you are? and a new startup that came around would say, we can do this too, we're embracing these tools, we're building now from the 2026 stack from the ground up, but they just have no idea who they are and they haven't developed that relationship with them over the years. How important is trust in this space? It's the most important thing. And I think that that is the area where, one of the areas where we've done really well as a team and as a business is, when we came out of the gate, And we sounded a lot more sober than a lot of other people who were pitching AI. So you sound serious then. And I think you said your co-founder has been at MSK for 15 years or something, right? And so they have a very different lens on these types of problems than probably other folks might have when they're just jumping into complicated bureaucracy at a government organization. That's right. And not to throw shade on the 20-year-old YC people, but we're not that. I've been in the industry. Ari has been at MSK. We understand the problems well. One of the things I've seen over and over and over again in my time in startups, because that's how I've spent my whole career, they've all been bio or life science startups, mostly in drugs and diagnostics. You get all these tech people who come in and think it's an easy problem to solve. And they're like, yeah, you show up, you do a little software, you do a little AI. How hard could it be? They're always humbled. And some part of me always feels gratified by that because it's like, there's all this hubris coming in. It's like, this is hard. Biology is really hard and it is far from a solved problem. And so it is important, I think, to have that perspective, that background, that humility coming in. And trust is exactly the word, Raj. I think it's so important. And that's what we're building at Weave as we build a company. It's not the product is a byproduct. We're just taking this technology, making it useful for people, and then they're coming to trust us. And that is ultimately the advantage that we have at this point relative to our competitors. And that's what our moat is. Training models doesn't really make sense anymore. It's passe, I think. You don't need to do that. There's other things to do. And it's a new regime. And trust is a really important ingredient in that new regime. Cool. I wanted to maybe zoom out a little before we move on to the lightning round. So I'm sure that this is in your pitch deck. It's been in many pitch decks before, but there's this idea called Eroom's Law, which is Moore's Law spelled backwards. And it's used to describe the slowing down and the increasing costs that it takes to bring a new drug to market. So I'm wondering if you could help us understand what fraction of the often quoted several $1 billion in a decade, some of the regulatory things that you're addressing at Weave. How much does it cost to submit an IND? That's not really something that I know. And how long does it take? How many man hours does it take to assemble that package of information? That's a great question. So, and I'll answer it on two levels. The first level is the immediate one. So an IND, if you go and outsource that, this is the easiest way to think about it, just have a consultant come in and write the whole thing for you. And this is after all the experiments have already been done. So you're not generating any new data, you're just preparing the package. That's going to cost you about half a million bucks just to produce the package of documentation. It's going to take you three to six months depending on how well organized you are, how much you know what you're doing, how relevant all the experiments are. An NDA, that market application on the other side, you're talking millions of dollars and a year or more to pull that information together. It's an order of magnitude or two bigger than an IND. So those are the stakes when it comes to just the paperwork part of this. But there's another way to think about it, which is why we picked this problem, why I'm excited about it. Regulatory drug development is regulatory development. They're synonymous. And so if you want to think about what in drug development is pertinent to the regulatory process, it's literally everything. And ultimately, the reason that you do any study you do, any clinical trial, any animal study, is because you're trying to improve your narrative. You're trying to improve your ability to get past the next regulatory milestone. And so you're only doing what you need to do to get past that milestone. So the whole thing literally revolves around this. And eventually, it's all about regulatory. And so we're starting here at the goal, the destination, is you want to get your IND done, you want to get your NDA done, and then working backwards into what should you do. And that is something that we'll be able to provide unique insight on. As time goes on and as our customers build more and more of their submissions within our platform, they'll have better insight into what works and what doesn't. And we will be able to provide better insight to the industry as a whole about what works and what doesn't and thereby make everything more efficient and faster. Got it. Thanks. So I also wonder if as you continue to grow and your platform's capabilities expand, what has to be in an IND is also changing. So the FDA previously said they would express willingness for computational models of antibody designs instead of animal models, which feels like a radically accelerating thing for FDA to be comfortable with. Do you see any other areas in the sort of total drug development and clinical trial pipeline that you think could be radically compressed? Like the benchmark is always like Project Warp Speed, where we went from nothing to shots in people's arms in about 10 months. Do you think that that will become more routine in the next five or 10 years? Or are there places outside of a global pandemic you think that we won't be able to meaningfully move the needle on? I think the thing that's really hard is ultimately you have to establish causality here. And there's just no way that I've ever been told about or thought of that you could do that outside of a clinical trial, a blinded trial of some kind. So you got to go to tests in humans. And that is just like ethically and morally challenging. You got to go about it the right way. It's very logistically challenging. And there's increasing demand for patients for clinical trials. So there's competition for patients for clinical trials. that I think that whole area, the clinical trials and how you design the studies, how you power them, how you recruit patients, I think is where more and more of the challenge is going to be. And then I think less and less on the biology side, not because it's easy, it's very not easy, but a lot of efforts have already been directed at that. And I think are pointing in the right direction and beginning to bear fruit, especially if you think about like all the omics data now that's out there, all that molecular data, I think is an essential ingredient in getting that part sorted. And then AI, I think is the next, the other piece that you need to make that part more tractable. So I think that's going to be in hand soon. Clinical trials, it seems like a tough nut to crack. Brandon, I think the thing that strikes me is that things that are amenable to brute force, either through measurement or through computation in drug development pipelines seem to be getting compressed. So like large-scale omics measurements, large-scale AI models are essentially brute force techniques. But I think I hear you saying correctly that there's no brute force approach for clinical trials. And so that's always going to be a fixed-in time and money cost for running trials. However, obviously other parts of the world like China to seem to be scaling clinical trials in a way that we seem to be unable to? Is that because of difference in regulation, difference in ethics, difference in centralized planning, or sort of like what explains the discrepancy between sort of those two clinical trial ecosystems? I certainly cannot profess to be any sort of expert on this, but what I've seen and heard is that, yes, the regulatory environment is more aligned to enabling that. And that is something that people in the U.S. are starting to advocate for and pay attention to. There are ways that we can change the regulations that don't meaningfully compromise safety, which is the main thing that FDA is trying to ensure here, but allow drugs to get to market much more quickly. Part of it is changing the way that the regulations work. Part of it is upgrading and evolving the processes associated with it across the board. There's just been this very concerted mindset, it seems, in China to orienting everything to this goal of creating new and better drugs and getting them to market quickly. And I don't think that there's any single silver bullet in that. It's a bunch of things that have to come together. I think that makes sense. and it's something that gets talked about in the Boston ecosystem a lot because there's a sense of competition, I think, with what's going on in China. So it's definitely a big topic of water cooler conversation here. Raj, do you think we should move on to the lightning round? Let's do it. Cool. Awesome. So these are simple rules. We'll ask you a bunch of semi-random questions. You can answer them seriously or unseriously, however you want. But I think we keep the answers brief in this section. How's that sound? Sounds good. Cool. All right. So the first lightning round question, I love this one because it's usually so revealing, is what was your first job? I was watering plants for a landscaper. And to the viewers who can't see his background, he was just telling me and Mike, the sound engineer that the plants in his background are ones that he grew from scratch. So it seems like a skill that you've continued to practice. Yeah, it's a through line. Nice. All right, here's our second one. What is your ultimate productivity hack? Calendaring, anything that you actually need to do. Having a to-do list. So budgeting time. Having a to-do list, a hundred things long does you no good if you don't have slots to put them into and time to actually get it done. Time is always the limiter. So I put stuff, I put all of I'm at a point now where I put everything in my to-do list on the calendar. And if it doesn't make it out of the calendar, it's not important enough. That's a good one. Cool. Okay. So here's a slightly different one. What's something that you'd never let an AI do, even if it could do it? Oh, that's a good one. Cook for me, I guess. I like to cook, garden, cook and garden. We'll say those two things. Those are things that are effort, but I think that the act of doing it is part of what makes it enjoyable, not just the result. Very cool. And I don't think we've ever asked that question on AI Grand Rounds. So it was a lightning round brand new one there and a good answer. All right, here's our next one. Who is your scientific hero? Or other than your PhD or postdoc advisors or, you know, undergrad or your master's research advisors, whose lab would you be in if you were in grad school today? Oh, that's a difficult one. I'm a big fan, and I got guff for this when I like wrote about this in grad school for some assignment once, but I'll say it again anyway. Still Carl Sagan, I would say, because of the communication, the narration, like the sort of for the love of science, not just like the act of science. You know, I think that he did a lot to open people's minds and inspire people. And I think that that's extremely difficult, especially for something that can often be so esoteric and complex and blah, blah, blah, and making that understandable and relatable, I think is incredibly important. So I don't know that I'm up enough to be transparent on kind of like who's doing what in academia. I've been much more focused on industry at this time in my life. So I don't know if I have a great answer for that. Maybe it's a cop-out, but I would go back to my old advisor, Ed, at UCSC, that one who got me started on this journey in the first place. I would love to work with and for him again, if that's the direction I was going to go. Yeah, plus one for Carl Sagan and his writing about the splendor of living on the pale blue dot. So no argument there. Okay, last one. I know that the beginning of the conversation, we talked about the importance of having domain expertise and like knowing an area sort of at a more macro level. Do you think progress in biology and drug development will be driven more by the biologists or more by tech and computer science people? My hot take on this is neither, I guess, neither and both. I think that like the new version of a biologist in five and 10 and 20 and 50 years from now is going to be an intrinsically computational and analytical person, more so than somebody who has kind of like deep knowledge of like a very specific area of biology and like wet lab skills. I think it's going to be people who are like able to work with and understand, manipulate large quantities of data and do that and use sophisticated tools to do that and marry that with like a sufficient understanding of the biology to know which questions to ask and know what makes sense and know what matters. It's interesting. We asked this question a lot and I think your answer is right in the sense that like it will be someone who has domain expertise, but that person, if we call them a biologist by today's standards would be like unrecognizable because the job will change so much. Yeah. Cool. Awesome. Well, Raj, how do you think he did? Did he pass? Totally pass. Flying colors. Nice. Congrats. All right. So for the last question, we usually like to zoom out a little bit and ask some bigger picture ones. It's interesting the area that you work in. It's one that Mark Andreessen has claimed will be one of the ones that is slowest to change, just because it's highly regulated. It's sort of designed to be slow by nature. So I'm wondering if you could give us a forecast for how you think AI in your area will grow, but then also how you imagine the larger AI ecosystem around you changing over the next five years. Yeah, absolutely. on a personal level the reason I got into this is the challenge I didn't want to do something easy I wanted to do something hard that mattered and in general I agree with that principle that it's a difficult area to work in it's conservative, it's slow moving but much faster than you might think and having been in this industry for a long time it's kind of shocking how much appetite there is across the board. I'm talking in FDA, in drug companies, in CROs for AI and all the promise that it harbors. So people want it, people are ready to do it. There's not the typical reactive barriers that are getting thrown up. There are perfectly reasonable skepticisms and a desire to understand and have proof, but that's different from don't even want to have a conversation. And at this point, to get a sense of how quickly things are moving, we have multiple customers now who have completed submissions that they built using our platform. So they used AI to produce the content for the regulatory submissions. They submitted those to FDA, they've been approved and those drugs are now in clinic. So that has already happened and not just for one, but for a number of our customers and more are coming. So people are already using this technology to produce submissions and use that to get drugs into the into the clinic i think that that pace is going to go faster going forward because i think and it comes back to the air rooms law thing everyone knows everyone understands that um we can't keep on the trajectory that we've been on otherwise we're just gonna get no more drugs into the market so something needs to fundamentally change. And I think people appreciate that this is the moment. I'm going to ask this question and then maybe one more, because I never like to end on this one, but over the same time horizon, what are you most worried about? If something were going to go really wrong, what do you think it would be? And I guess, what keeps you up at night? Most of my concerns are much near term and for my business and like growing it and helping it survive to be perfectly transparent. Maybe I'll go out kind of like a different direction on this, because I think I was just reading the New Yorker article about Sam Altman and a lot of the concerns, I guess, that like he has and others have about AGI and so forth. I'm just like a very skeptical person by nature. And I've been called, I think myself as a realist and I've been called like a pessimist and I don't mind that. So just to give you a sense of kind of where I'm coming from. I don't think, I'm not really worried about AGI. I think the LLMs are amazing, powerful technology, but it's got a lot of constraints and flaws that when you work very closely with it, like me and my team do, you see those like up close and personal and you're intimately familiar with them. So like, it's a very powerful technology. It has to be wielded by people. And I think that that's going to be the case for the foreseeable future, maybe for the distant future. But I don't think I'm not of the belief like a lot of people in the Bay and in this industry that, you know, AGI is right around the corner. So I don't really fret about that too much. I'm just more excited about the potential of the technology. Yeah, I mean, I kind of joke, but I say that like, they don't let you on the BART unless you say a recursive self improvement, the doorstone, you know, so like, that's a very measured take for someone who spends most of their time in the Bay. Thank you. I try. Yeah. The final question then is like the flip of that. If we can, if you come back on the podcast five years from now, what will you have hoped to have seen both sort of at Weave, but more broadly in the AI landscape? I think on that time scale, AI is going to become boring. And I'm excited about that because that's when things really begin to change. You know, the internet didn't happen in the dot-com era. That was just the first phase. And there's all this excitement and hype. And then there was a crash. And then now we have the real internet, which is like the, you know, social networking and Amazon and, and all of the stuff that was built on that, which is incredible. It's completely changed the way that the world works. But we don't talk about the internet like it's a great, cool, exciting thing anymore. It's just there. And so I think, and I hope, that in five to 10 years, that's where we're going to be with AI and kind of through this frothy phase of the investment environment is overheated, in my opinion. And there's going to have to be some sort of rectification there, and that's fine. And once that happens, then like we can get to the business of, we can get down to it and figuring out how are we actually going to use these things to change what we're doing in a meaningful way. And that's just, there's no way around that being a process that takes years. And I think we're at the beginning of that now and we're going to really see the fruits of that in the coming years. Awesome. Thanks. Well, I think that that's it for today. Thanks for coming on the podcast, Brandon. My pleasure. Thanks for having me. That was great, Brandon. Thanks so much. Yep. This copyrighted podcast from the Massachusetts Medical Society may not be reproduced, distributed, or used for commercial purposes without prior written permission of the Massachusetts Medical Society. For information on reusing NEJM group podcasts, please visit the permissions and licensing page at the NEJM website.