Drug development is an incredibly long, complex, and expensive process — over $2 billion and 10 years — with most medicines failing late in clinical trials, making it harder to send things to space than to get a new medicine approved.

(00:00:06) Hi listeners, welcome back to No Priors. (00:00:08) Today I'm here with Saji, the co-founder and CEO of Benchling, the system of record for biotech R&D. (00:00:14) Today we talk about the state of AI in bio, Benchling's bet on (00:00:20) AI agents to help scientists make better decisions, experiment faster, and deliver drugs more effectively. (00:00:27) Why drug programs are so expensive and fail so often, and how to build a culture of science and software together. (00:00:35) Saji, thanks so much for being here. (00:00:36) Thanks for having me, Sarah. (00:00:37) Excited to be here. (00:00:38) Okay, so for our general listener base, can you just give us an overview of what Benchling is and sort of the scale of the business today? (00:00:45) Sure. (00:00:45) I'm one of the co-founders of Benchling. (00:00:48) We make modern software for scientific progress. (00:00:50) So I started the company about 13 years ago. (00:00:53) It's been a long time. (00:00:54) Oh my God. (00:00:55) I know. (00:00:56) So I'm a software engineer by background, but I worked in a biology lab. (00:00:59) I was like really interested in medicine and coming from the world of software and software developers have amazing tools for working on code and for collaborating. (00:01:07) And when I got to the biology lab, I found that scientists had paper notebooks and spreadsheets that would sit on their desktops and like (00:01:16) It was terrible. (00:01:17) And so it was really hard to work together. (00:01:19) And I think that was really frustrating for me personally. (00:01:21) And I thought, a little bit naive at the time, I thought, how hard would it be to build good tools for scientists? (00:01:27) And so I started working on Benchling, which helps scientists design molecules, plan their experiments out, run those experiments in the lab, get the data, organize it, analyze it, and then share it with their colleagues. (00:01:37) Today we work with about 1,300 biotech and pharma companies, scientists at over 7,000 academic (00:01:44) institutions, universities all around the world, and our software powers, household names like Moderna and Sanofi and Eli Lilly and Regeneron, but also like cutting-edge biotech startups, the... (00:01:57) future AI biotechs like Isomorphic Labs and Zara and companies like that. (00:02:02) So we get to see the innovation happening across the entire biotech sector and build software that helps power it. (00:02:08) I'm super excited to like actually use that vantage point and ask you a bunch of questions about bio in the macro, but just so people who don't come from the domain can picture it a little bit better, I think like, you know, I can picture like gene sequences and like the assay like said yes or no, like what other types of, what is the (00:02:27) data that's actually inventionally. (00:02:28) I think what's really interesting for everyone to understand is like making a drug, there's like 9,999 steps in making a drug after you come up with a molecule. (00:02:37) So you have to make a medicine, you have to find a biologically meaningful target in the body, something you want a drug. (00:02:45) You have to design A molecule to optimize that molecule. (00:02:48) You have to test that molecule in Petri dishes and cell lines and (00:02:52) animals, various kinds of animals. (00:02:54) Then eventually you get to the point where you can take it to a clinical trial and you're testing it in subsequently larger groups of humans. (00:03:00) All the while you're figuring out how do I manufacture this thing and develop a process to make it scale economically, safety with the high, with quality, all while navigating regulatory bodies. (00:03:11) So that eventually in 7 to 10 years, you can have a drug that you give to people commercially. (00:03:16) And even then there's still more work there. (00:03:18) So it's just incredibly long and complex process. (00:03:21) And where Benchling focuses (00:03:22) is all of the scientific data that comes out of the lab. (00:03:25) So everything from all the different types of molecules that are being created, to how they're related, to the work that went into creating them, to the different types of tests that you're running on them, to the data coming back from the animals, to the, you know, scale up data coming out of the fermenters when you're figuring out the process to manufacture it. (00:03:40) All of that incredibly rich and heterogeneous scientific data has to be brought together in one place, organized, made searchable so that scientists can make decisions based on it. (00:03:49) If we go zoom out for people just (00:03:52) just like looking at biotech from the outside, it seems a very macro sensitive industry, right? (00:03:58) And we are perhaps coming out of like kind of an ugly period. (00:04:02) Can you just characterize like where we are in the bio macro cycle? (00:04:05) Yeah, And I'm definitely not a sort of macro specialist, otherwise I'd probably be, you know, an investor or something like that. (00:04:10) But it's all your customers. (00:04:11) Yeah, it is. (00:04:13) I would say like biotech has, it is definitely an industry that has gone through cycles. (00:04:18) We're probably like the last couple of years are probably like the equivalent of like the (00:04:22) Because it's pretty bad. (00:04:24) Biotech. (00:04:25) Yeah, it's been a tough time. (00:04:27) COVID was sort of the peak when mRNA was this thing that kind of like reopened the world. (00:04:32) And there's a lot of generalist money that came in and a lot of exuberance and excitement. (00:04:37) And it's not, you know, the sort of dot-com bust equivalent wasn't just because of that. (00:04:40) There was changes in interest rates, tariffs, regulatory uncertainty, China, a bunch of different factors and some including like scientific technologies that we got really, really excited about that are still very important (00:04:52) and promising, but maybe haven't become commercially successful as fast as people wanted. (00:04:56) So a whole confluence of factors there. (00:04:58) What are you referring to in terms of scientific knowledge that got people hyped? (00:05:03) Yeah, like I would say there's a lot of generalist excitement for gene editing, cell and gene therapies, RNA, and all of these are- So new delivery methods. (00:05:10) Yeah. (00:05:11) New, like, I would call them kind of categories or form factors of medicines, modalities is the word, but like, you know, the last decade actually, maybe even longer of biotech has really been this story of (00:05:22) of new categories of medicines being sort of invented and taken to patients. (00:05:28) And some of these, like, there are approved gene editing medicines. (00:05:31) There are approved cell therapies where you're reprogramming the patient's immune system. (00:05:34) There are approved gene therapies. (00:05:36) There's approved mRNA medicines. (00:05:38) So these are real categories. (00:05:40) But I think investors and companies got very excited and put a lot of money into these categories. (00:05:47) And we're kind of in the trough of disillusionment for some of them now. (00:05:50) And my understanding is that they have taken (00:05:52) taken longer and been more expensive than people expected, than investors expected. (00:05:57) Absolutely. (00:05:58) And in 2021, every biotech was getting told by investors, like, you need to build a platform company that's going to cure a bunch of different diseases. (00:06:05) And here's hundreds of millions of dollars and capital is free. (00:06:08) And investors can change their strategies a lot faster than companies can. (00:06:12) And a lot of those companies, because they were taking such a big risk on like a new form of technology, we're going to be the next RNA company, next cell therapy company, they picked (00:06:22) diseases that might have been like simpler problems to solve with smaller patient populations. (00:06:26) And then all of a sudden, investors change their minds. (00:06:29) Platforms are no longer valuable. (00:06:30) You're working on a thing that like is just supposed to be a proof of concept and all of a sudden it's like defining you. (00:06:36) And so it's a really tough spot for those companies to be in. (00:06:38) What's the relevance of China in all of us? (00:06:41) I think so if the last decade was about sort of biologics and these new modalities, I think the next is going to be about speeding costs. (00:06:48) Like people want more drugs and they want them cheaper. (00:06:51) And China, (00:06:52) is very good at things related to speed and cost. (00:06:54) And so all of a sudden in the last couple of years, you've seen this rise of Chinese biotech companies that are able to create molecules and bring them to patients in clinical trials in China, just our early phases of clinical development, really fast and really cheap, even in some of these new modalities. (00:07:09) And so you've seen this huge uptick in pharma (00:07:12) going to China and buying molecules that they typically would have bought from American biotechs. (00:07:16) And this is like top 30 pharma. (00:07:18) Yeah, These are the biggest companies in the world, the Mercks and Pfizers and Lily's and so forth. (00:07:25) Many of them have like gone to China and bought molecules that historically they would have bought from American biotechs. (00:07:30) Are there medicines that people would recognize like in market today? (00:07:35) One of the most notable medicines that people might recognize is called Carviki, and it's a Johnson & Johnson medicine. (00:07:42) So Johnson & Johnson partnered with a Chinese biotech called Legend Biotech. (00:07:46) They saw the data that Legend presented, and it was a time when people were pretty suspicious. (00:07:51) And so they were like, it's like, the data's probably not going to replicate. (00:07:54) It might not be real, but like J&J, I think, saw it and realized how promising it was, and they've taken it. (00:07:58) And it's actually a cancer immunotherapy. (00:08:02) So it kind of reprograms the human immune system. (00:08:03) I think it's for (00:08:05) multiple myeloma. (00:08:07) And like that medicine is very commercially successful and widely distributed in the US to those cancer patients. (00:08:13) What's been the reaction of like Western biotechs to this? (00:08:18) It's a mixed bag. (00:08:19) I think there's some folks who are like, it's kind of inspired that American biotech needs to be faster, cheaper, more competitive. (00:08:27) There's some more nationalistic reactions, I think, of like, hey, like, well, there's different regulatory or ethical standards over there. (00:08:34) Are the data, are they all going to replicate? (00:08:36) So some skepticism as well. (00:08:38) But by and large, I think it's like very much here to stay that China's going to be like a major, major biotech player. (00:08:44) Yeah, we can spend this whole time talking about macro. (00:08:46) Yeah, I want to get to the meat of our discussion, which I also think is, there's some premise that like the answer to faster, cheaper, better might in part be AI in biotech. (00:08:58) Yeah, I think I believe that now. (00:09:01) And it's really interesting to see like the general public, you know, big tech, startups, the model labs, everyone is saying like AI is going to cure a disease. (00:09:09) So it's very good that everyone's excited by that. (00:09:12) You, I don't think of you, I think there's an amazing CEO, but not really a content marketing guy to date. (00:09:17) And you wrote an essay very recently that I thought was amazing about how we can possibly change like the scientific field in biotech with AI. (00:09:27) Can you give us the cliff notes on it? (00:09:28) And then we'll link it in the show. (00:09:29) Absolutely. (00:09:31) Yeah, I think like maybe to step back, like one thing I just like (00:09:36) wish people would appreciate more is like medicines are, medicines are magic, I think. (00:09:41) like we take for granted how awesome medicines are. (00:09:45) I think 9% of healthcare spent, prescription drug sales are 9% of healthcare spending in the US. (00:09:50) Like we have obviously this healthcare cost problem, but drugs are this amazing ROI. (00:09:56) And the best part about drugs is they go generic. (00:09:58) So a drug today is only going to get cheaper over time and it works just as effectively. (00:10:03) I take a statin today that probably costs like nothing. (00:10:06) And 20 years ago, it was some expensive medicine. (00:10:09) And that's like- It's not obvious any other part of the healthcare system gets cheaper over time. (00:10:13) It's not, yeah. (00:10:14) The rest of healthcare is very labor dependent and labor generally gets more expensive over time. (00:10:18) I am very optimistic for AI to help there too. (00:10:20) But drugs are this amazing thing. (00:10:22) We should want more of them. (00:10:23) And then we get to like stockpile more and more of these amazing medicines. (00:10:26) But it takes over $2 billion, generally about 10 years to bring a medicine to market. (00:10:32) And most of those medicines will fail very late in this process. (00:10:36) You get 7 to 10 years in, you've spent hundreds of millions of dollars, and clinical trial fails. (00:10:41) Medicine's not safe or not effective. (00:10:43) And so it's an unbelievably like difficult pursuit. (00:10:46) It is probably easier at this point to send things to space or to put people on the moon than it is to get a new medicine approved. (00:10:52) And I know $2 billion probably isn't that, I feel like AI has desensitized us all. (00:10:58) everything is like $100 billion data centers and whatever, like $2 billion, like what's that? (00:11:03) But when there's that high of a failure rate, it's very difficult for investors to underwrite that. (00:11:08) And that was, while we had all these new categories of medicines being kind of invented over the last decade, I think that's like, that's important and it's here to stay, but like the industry has to change, like the pressure on biotech to be faster and cheaper is just (00:11:23) higher than it's ever been before. (00:11:24) I think a lot of that cost comes from how artisanal the industry is. (00:11:29) Like biotech is this place where if you look, I'll sort of take the digital and physical realms for a second, they've actually done a good job of systematizing the physical realm. (00:11:38) You brought up sequencing earlier, like Illumina has put sequencers on every single bench in every single lab, and now sequencing is this accessible tool to all of science. (00:11:47) You could say the same thing has happened with different like reagents and lab consumables and things like that. (00:11:52) But (00:11:52) If you look at the digital realm, where it's like how people collaborate, how data is structured and shared. (00:12:00) the workflows that are used in science, which is all about collecting data, all of that's basically bespoke and invented one-off by every company. (00:12:09) It's because those companies are playing kind of a sort of a one-time game because the process is so long that you're sort of just trying to survive until you get six, seven years in, you show some clinical success and a pharma company comes and buys you. (00:12:22) So you're not really like building for scale and building for durability. (00:12:25) That seems like it also comes from some of the structure of where the innovation happens, right? (00:12:30) Because if you were doing it across a whole portfolio and actually starting at 0 and you owned the innovation, then you would invest in the systems. (00:12:37) Totally. (00:12:38) Yeah. (00:12:38) If you were setting out to build a company that (00:12:41) was going to, you wanted to build the next great pharma company and have a whole portfolio of medicines, you probably care a lot about that, but that's such a, that's like a high capital, long-term, high-risk thing to do. (00:12:50) It's very hard. (00:12:50) And after seven, eight years, and you have some good clinical data, like, do I roll the dice again and keep going for another 10 or do I sell? (00:12:56) So I think like, because it's so artisanal, there's this huge opportunity now with AI to get more shots on goal, faster, cheaper, make better molecules, (00:13:06) and then bring them to the clinic safely and faster. (00:13:10) And I think that's the big opportunity. (00:13:12) People get very focused on clinical trials because they're like the biggest line item. (00:13:18) And they're important, don't get me wrong, but I think it's actually a bit of a red herring where, yes, there are operational problems, like some studies are designed badly, it's hard to recruit patients. (00:13:27) sticker price is really big, but at the end of the day, a lot of molecules are just not good. (00:13:33) And so we need better molecules and we need to move them to people faster. (00:13:37) One other. (00:13:39) criticism that you kind of imply in your essay as well of like why the industry isn't more efficient is that even the large pharma companies are not as good at buying innovation and finding it as they could be, right? (00:13:53) And so examples of GLP-1s and Contruda, like some of the amazing breakout successes were not super obvious to the buyers. (00:14:02) Yeah, I think those two stories are really interesting. (00:14:05) There's a great quote from Dario, the Anthropic CEO, and his kind of essay about the returns to intelligence in scientific progress are very high. (00:14:14) We're talking about machines of love and grace. (00:14:16) Yeah. (00:14:16) So the returns to intelligence are very high. (00:14:18) And I think like the stories of GLP-1s and Keytruda are like great, great examples of that. (00:14:23) So GLP-1s obviously have just transformed obesity as like a treatable disease when, by the way, it was like a totally unfundable category of things like five years ago. (00:14:31) Why do you think it was unfundable? (00:14:33) I think, like, again, because we know so little. (00:14:35) biology and there are so many failures in that space. (00:14:38) And again, running a clinical trial for obesity where you need huge populations of people that you monitor for very long periods of time, like super, super expensive and everything has failed before. (00:14:48) Like pharma companies generally aren't willing to underwrite that stuff sometimes. (00:14:51) I mean, neurodegenerative diseases are the same way, like Alzheimer's is just like graveyard of billion dollar failures and like it's getting back in now, but there's a period of time where everyone left the space. (00:15:02) And so, but the core science for GLP-1s was (00:15:05) kind of sitting on the shelf in some sense. (00:15:07) Like it's been known since like the 90s. (00:15:10) And so it took some insights and conviction. (00:15:13) And then all of a sudden, like we have this category defining medicine that's going to go on to probably be the best selling drug of all time. (00:15:19) And that's happening. (00:15:20) And then Keytruda is a similar story where there's a molecule that's gone through a couple different acquisitions. (00:15:25) And it's almost like it's at the bottom of some list to be like out licensed and sold off. (00:15:30) And then a competitive thread pops up and someone sees that, hey, this is kind of like Keytruda. (00:15:35) And so (00:15:35) So, and credit to Merck, they had the courage to go all in after they realized what it could be. (00:15:41) So just another example of like, there's a lot, it's a pretty inefficient system. (00:15:46) And people are pretty, they're rational actors. (00:15:48) It's just that we don't know a lot about biology and our ability to predict what's going to happen in the clinic is so poor. (00:15:55) And the cost to get there and to make those decisions is so high. (00:15:58) And so if you can get to the clinic faster, cheaper, like failure in the software world is you work on a product for a year or two, you spend a couple million bucks. (00:16:05) and it doesn't work, but in biotech, you're underwriting four, five, six years, big team, hundreds of millions of dollars. (00:16:11) So how do you compress that so you get feedback faster? (00:16:14) So Benchling is a system of record company. (00:16:17) It's a data platform. (00:16:18) What is Benchling AI? (00:16:20) Benchling AI has kind of two major components to it. (00:16:23) The first is tools for simulation. (00:16:26) So this is taking open source, proprietary companies' internal models and making them accessible to scientists directly in their workflow. (00:16:34) So the right model at the (00:16:35) the right moment in the scientific workflow, already set up so that a wet lab scientist without computational skills can use it effectively. (00:16:42) And then the results are linked to all of their other information in Benchling. (00:16:47) And then we also see that... (00:16:49) laddering up to being able to help scientists recommend, like help recommend for scientists the next best experiment to run based on all the work they've done in the past, plus all the public literature available. (00:16:59) And so we think it's like an exciting way to approach the co-scientist problem. (00:17:03) Then the other facet of benchling AI is agents that automate work for you. (00:17:07) And so we've released this deep research agent. (00:17:09) It works similar to the deep research agents from Anthropic and other foundation labs. (00:17:14) But what it does is it works over benchling data, (00:17:17) with the context of the benchling data model. (00:17:20) And so it enables scientists to ask these very difficult, and science is fundamentally about like asking and answering questions. (00:17:25) And so for our customers, it helps them to do a type of question that previous in the past would have taken weeks or months to do and do that in just a couple hours. (00:17:34) So a great example of this is with a customer that was getting ready to run some mouse studies and they were looking at 20 different mouse models and they used a deep research, our deep research capability to look at all the historical (00:17:47) of mouse studies that they had run. (00:17:48) And it turned out that a bunch of the mouse models that they were about to investigate, which would have taken eight months, huge cost, big experiment to run, someone had already done before. (00:17:58) And it was trapped in some lab notebook from many years ago from a company that had been bought. (00:18:04) And all the people were long gone. (00:18:05) And so there's so much of science that lives in folklore and institutional knowledge, and that's just kind of lost over time. (00:18:11) And so we sort of view this as being able to unlock memory for these organizations and help make scientific data reusable over time. (00:18:17) just accelerate because they didn't have to do that piece of experimentation anymore. (00:18:21) Exactly. (00:18:22) And so we're working towards a world where like there are AI agents that can do all sorts of different tasks in the scientific process, whether it's generating reports and asking questions, or it's even like composing experiments from while you're in the lab with voice and vision and things like that. (00:18:36) If you project out a few years, like everybody loves to talk about this idea of like the AI scientist, a lot of autonomy, AI co-scientist. (00:18:44) What do you think is the role of (00:18:47) scientists like a couple years out. (00:18:49) Oh wow, that's so interesting. (00:18:50) So yeah, when I hear from AI scientists, I think it definitely evokes this image of a kind of fully AI-ified design, make, test, analyze, loop. (00:19:00) And we'll sit back and let the robots give us drugs. (00:19:03) And while I would love for that to happen, and I'm maybe more optimistic on a longer time scale, we will get there. (00:19:10) I think in the short term, I'm next one to two years, which already feels like an eternity in AI time. (00:19:16) I'm a little bit more (00:19:17) more bullish on sort of the augmentation model. (00:19:19) Like I kind of think of it as like a Waymo versus Tesla approach, where you can do the Waymo approach to autonomy. (00:19:26) You just need a lot of money and a lot of patience. (00:19:28) And it's going to take some time. (00:19:30) I think the Tesla approach has been a little bit more, I would say, (00:19:34) taking steps. (00:19:34) I don't want to call it incremental because it's not. (00:19:38) And so I think if you can kind of get those ingredients, take the Waymo approach, which some companies have, that's awesome. (00:19:43) But I think for the rest of science, there's a huge opportunity to just like make things better one experiment at a time and pick off a lot of low-hanging fruit and see if we can get 7 to 10 years down to two to three years and a lot fewer specialized roles and a lot cheaper to bring a drug to market. (00:20:01) I think actually like radiology is like an interesting parallel where I feel like ML people have been saying radiologists are going to go away for 10 years. (00:20:09) But I think the model that's worked there. (00:20:10) I think like 40, yeah. (00:20:11) Probably. (00:20:12) I think the model that's worked there though is like kind of the copilot model. (00:20:16) And truthfully, like at the end of the day, you probably like, you know, with a radiologist, you probably need a human to be accountable for those decisions. (00:20:23) It's not just about the technology. (00:20:25) Like someone's got to be there to like, I don't know, get sued if something goes wrong. (00:20:30) Yeah. (00:20:31) I mean, that makes sense to me in clinical practice. (00:20:34) I'm more hopeful that like some of the experimental decisions can be more automated. (00:20:40) But one question that I think biology faces that other fields in AI face as well is the question of like, (00:20:49) how do you make these agents like useful, transparent to specialists outside of the domain, right? (00:20:56) So if you think about engineers generating a ton of code, like there's a lot of looks good to me, I didn't really read it. (00:21:02) I don't know if that's a good architectural decision, like what's happening. (00:21:05) How do you think about that for like, for example, wet lab scientists and (00:21:11) computational analysis, they don't necessarily like deeply grok. (00:21:14) Yeah, I think right now, when I look at biotech, we are in, so that's absolutely like the right point of like, are scientists going to trust this? (00:21:21) And how do we know if it's accurate? (00:21:23) Right now, I would say like, there's been amazing (00:21:28) advances in capabilities that scientists could use in the life sciences from the foundation model labs, from bio AI companies, from everyone. (00:21:36) It's really awesome. (00:21:37) But I think we're like, we've got GPT, but there's no chat. (00:21:40) That's kind of how I think about it. (00:21:42) I think. (00:21:43) The chat, and I mean chat metaphorically, like that was the interface that made things really take off in software. (00:21:50) And I don't think it's like really, we haven't figured out what that is in bio yet. (00:21:53) There's some ideas, but by and large, and I just got back from a month on the road and I was in Boston, London, bunch of other places that are sort of scientific capitals outside of SF. (00:22:03) And like most people aren't really using that much AI and R&D yet. (00:22:08) They all want to, they're primed to, but there's a lot of concerns about (00:22:11) accuracy, IP, security, legal. (00:22:15) And I think the farther you go from SF, the like larger those concerns get. (00:22:20) And so you're optimistic that you can make a lot of the like context, value, whatever is important for scientists in different domains to understand about an output like legible through the product itself. (00:22:34) Yeah, I think that's like, I think- Legible enough to be useful. (00:22:37) Yeah, I think in a vertical, I think 90% of the work is actually like transition. (00:22:42) It's taking something and making sure scientists trust it. (00:22:45) It's the right point in their workflow. (00:22:47) It's easy to use, and it's accurate. (00:22:49) I think the AI that wins is going to be the one that people actually use. (00:22:53) Give us the temperature check of like what large pharma and your customer base thinks about AI right now. (00:22:58) They've got these AI officers. (00:23:00) Oh yeah, there's excitement for sure. (00:23:04) There is optimism and belief. (00:23:06) I think they're pretty pragmatic though. (00:23:08) And I think they're all looking to transform, but they're being methodical. (00:23:14) Like I would say most of the (00:23:17) large pharma at this point that I've worked with, like, they've got copilot and things like that. (00:23:23) And they're doing a lot of pilots of different technologies, but I haven't seen their R&D orgs transformed yet. (00:23:28) Now, the one place I would say that pharma has really leaned in and has an advantage is they have incredible data generation capabilities. (00:23:35) And so many of them can and should be training models. (00:23:38) Like experimental data generation. (00:23:40) They can generate data to train their own models at a scale that most biotech startups can't match. (00:23:46) So I think while it's (00:23:47) early on sort of the agentic how we work side, I think you're going to see very unique models come out of pharma where their computational scientists are building interesting predictive models that, you know, similar to what's happening in the open source world.