Rob Toews, Radical Ventures Partner on Investing in Frontier AI

Speaker A: Oh, one sec.

Speaker B: I don't know.

Speaker A: Why is there like a, uh, technical. Oh, we're live. Okay. Hi Rob. Welcome to Venture with Grace.

Speaker B: Thanks for having me. It's great to be here.

Speaker A: Uh, before we start our show, I want to give a quick shout out to our amazing sponsor. This episode is brought to you by Navias, the ultimate cloud for AI, uh, innovators. Navius provides AI infrastructure. You can combine reliability and speed with flexibility and, and engineering support are matched by hyperscalers. AI leaders like Meta, Shopify and Higgs Field already partnered with Navius to run their AI workloads. Plus, venture backed startup can save up to 150k on compute costs when they apply for access. Visit navius.com or navius.com startup to learn more. Okay, um, before we start, I want to say I believe like Jay from 12 Labs said amazing things about you guys on our show. So that's like when I first heard about like you and Radical Venture, and then it was so amazing to meet in person at Human Acts. And um, I want to start with like, you know, you started your career, I believe it was at Highland Capital and in the Bay Area, and before that you were in consulting. Maybe we could start with like, what were some core lessons that you've learned early on in your career, um, that kind of shaped you into who you are today?

Speaker B: Yeah, totally. First of all, it's great to be here and uh, uh, we Love Jay and 12 Labs. They're the best. And it's, it's been a privilege to partner with them from their very earliest days. Um, yeah, maybe just as quick, quick kind of background and context. I started my career at Bain Co. And consulting for a few years. I did grad school at Harvard, I did the J.D. mBA there, so Law, uh, school and business school. And then I worked in autonomous vehicles. For a few years I worked at Zoox, which was an autonomous vehicle startup. Um, I spent a little bit of time on the policy side of things. I worked in the White House during President Obama's administration working on autonomous vehicle policy. And then for the last eight years or so I've been a vc, totally focused on AI. As you mentioned, I started my VC career at Highland where I was the kind, um, of the point person for AI. And then for the past five years or so I've been at Radical where I lead their San Francisco office and have led a number of our investments. And so, yeah, to your question, around kind of big picture lessons, I would say yeah, it's been kind of A diverse um m Kind of um multi layered set of career stops from consulting to policy to startups to vc. And I like there's a lot that's different about those different roles but there are also some like core fundamental uh, consistencies and insights. I think one. And these all to some extent sound cliche but I think also have like profound truth to them. Um one is just that people matter more than anything and the quality of people matter more than anything. And I think I saw that in, in the world of startups like the reason Zoox was so incredibly successful in the, in the window that I was there was uh, it was just such a talent magnet and like the very best people came there and I think if you know looking at talent flows and startups is so informative because at the end of the day wherever the best people are going is going to be successful. Um, and I certainly that's true today in venture. Like I feel like the more years that I spend in venture the more I come to appreciate that like really matter. What matters so much more than anything else is just the quality of the people the founders. Like if you back the best people good things will happen and like you can, you know, you can get caught up in the TAM and the gross margins and the growth and the competitive landscape and of course all those things are relevant and you should pay attention to them. But at the end of the day like it really is just people, people, people. And if you just. Actually I saw quote Arthur Rock who's one of the like the original venture capitalists, um, or early investor in Apple and Intel and others. Um, he, he summed up like the lesson of venture capital in four words which is back the right people which I think is just like completely, completely true. So anyway I think that's one of the most important lessons and then a second lesson that I would share that I think I've come to appreciate more and more is um, like intellectual curiosity pays such huge dividends over time and if you put yourself in a role where you're genuinely, you genuinely enjoy what you're doing and you are like excited and kind of internally motivated to learn more and to dig in and ask questions like you just are going to be successful versus doing something that you don't feel that internal passion about. Um, and so I think like yeah, aligning your, your genuine internal interests with what you're doing day to day just is such a recipe for success for sure.

Speaker A: I totally agree with you on like um, essentially like spotting all the talent really following where the top like talents are going is super important. Um, I feel like you have like such a great like AI portfolio, whether it's like Crusoe, like Chase spoke at our conference and our podcast, um, as well as like you know, writer, um, hippie at like a lot of like or generalists in the physical ad space. Um, you have invested a lot of like really great companies. Um, why don't we start with like you know, um, the investing strategy of like Radical and obviously you guys have like an early stage fund and a leader stage slash like grow stage fund and um, leading the Bay Area or like west coast, um, team like how does it kind of like came together and maybe we could talk about like the day to day of like um, thinking about the, whether it's like investing strategy and the, the newest like AI trends as well.

Speaker B: Yeah, totally. Yeah. Maybe just to give some very quick background on Radicle as a firm. Um, we are a venture capital firm that's entirely focused on AI. Uh, Radicle is one of the oldest and one of the largest VC firms that are, that's totally AI focused. So it's kind of one of the kind of OG AI oriented firms. Um, we have our roots in the early deep learning research ecosystem. Um, and as a result of that have close relationships with a lot of the leading pioneers of AI and also with the kind of the AI research ecosystem more broadly. So uh, we're very close with people like Jeff Hinton and Fei Fei Li,

Speaker A: um, who are partners LPs as well. Right.

Speaker B: So yeah, LPs of ours. Fei Fei is a partner at Radical, um, and yeah, we work very closely with them. And uh, those relationships have been really valuable for us as we've grown super quickly. So the firm was founded in Toronto which is where I think people who are deep in AI know that that's kind of where modern AI was invented out of Jeff Hinton's lab. And Ilya, who would go on to be the co founder of OpenAI was Jeff Student and they kind of had this big breakthrough in 2012, um, called uh, Alexnet. Um, so anyway, so radical kind of came out of that ecosystem. But today we have offices in San Francisco, in New York, in Toronto and in London. Um, we have just under $3 billion in total under management. As you mentioned. We have, we're investing out of two different uh, strategies currently. An early stage fund and a growth fund. The early fund is a $650 million fund. The growth fund is an $800 million fund. Uh, out of the early fund we basically do series A and earlier but we love going very very early. So we will frequently incubate companies and get companies started right at the beginning. Um, we did that with World Labs as one example of fei phase company. We did that with Cohere. Cohere was actually the very first company we incubated back in 2020. Uh, more recently we incubated Emerald AI and Reka and a handful of other companies. So we like getting involved very early. Uh, and then on the growth side we invest uh, really series B and beyond. Um, and that includes more mature later stage businesses like Crusoe that you mentioned and others. Um, and yeah, and I think like uh, AI is a globally relevant and globally impactful technology and so I think it's a huge advantage for radical that we are a really global firm in our DNA. We have a big team in San Francisco, we have a big team on the east coast between New York and Toronto. We have a very strong group of folks in London as well. Um, with close size of DeepMind and I think London in particular is such a massive hub of AI talent and startup activity largely because DeepMind has kind of over the years like aggregated all the best talent across Europe and Israel into this one company. And of course a lot of inevitably a lot of people are leaving leave DeepMind to start companies. And so we, one of our partners based in London was at DeepMind for several years before joining us and he has just done an amazing job of backing a lot of the best founders coming out of DeepMind starting companies. So anyway it's yeah it's been a lot of fun.

Speaker A: I love it. I just saw like outside AI also AI was ah, featuring our like smart AI 100 list. So I launched a list of like CU series a company almost every week there's like a new round sets like raised from the list and um, all set was one and like I just saw um and then the founder was on our podcast as well and as well as like you know I saw you guys invest in u.com like the CEO was on our podcast as well and obviously I feel like you guys have invested like some of the most like iconic companies like Word Lab and then I know that we um, um are like we. So I do like a weekly of like announcing like the new rounds list so D card recently raised as well. So maybe we could start with like you know in the west coast like since you mentioned like a lot of the AI talents are coming out of like different labs and obviously you guys work very closely with like many partners that's like the OG in the industry. Um, can we talk about like maybe the talent shift? Like so I saw a lot of people joining open like whether it's like anthropic or other companies are like coming out of like let's call it like Stripe or other, you know, uh, AI unicorns. Like where does the top AI talent typically come from these days? And because of today every company is an AI company. Like what would be the, I would say like you know, investing strategy compared to let's call it like three years ago.

Speaker B: Yeah, yeah, no great questions to the first question in terms of where AI talent is coming from. I think it's helpful to kind of like segment AI talent, uh, and also AI startups into a couple different categories. There are like research first AI companies that are like doing genuine AI research, developing novel architectures or pursuing novel modalities. You know it does obviously doesn't have to be in language, could be you know, video AI or uh, AI in biology or in chemistry. Um, but there is still, despite the fact that the AI has kind of taken over the world and it's, it's on the tip of everyone's tongue, there still is a relatively small number of individuals globally who, who have genuine legitimate talent building frontier AI systems. And you know it's we, I used to say it's like under a thousand people, you know, maybe it's like a few thousand people now, but it's certainly less than 10,000 people globally. And you know, most of those people are at Google, are at a few organizations Google or OpenAI or Anthropic or Meta, uh, you know, a handful of startups that are sort of frontier labs. And obviously there's, there's been this new movement of like NEO labs that are being founded, um, but largely by alums of those organizations. So it is still a pretty small subset of organizations that produce real frontier AI research talent. Uh, and it's kind of a joke but the fact is it's kind of true. People don't really leave anthropic. Very few people leave anthropic and start companies. So it's largely Google DeepMind as well as OpenAI and Meta. Um, and you know, people have been leaving XAI recently. Um, so but that's all, that's kind of like the AI research side. Uh, but that's only one archetype of interesting AI startup. There's obviously like so many opportunities to build applied AI companies that don't require you to be doing AI research training your Own models from scratch, so on and so forth, um, that still require like I think there's still massive advantages to having hardcore technical talent, um, that is deeply familiar with AI technologies and models and is AI native. But um, that's a much broader pool of individuals and you can see those founders coming from anywhere, including being really young founders that don't have much of a professional track record at all. And this is like a increasingly common archetype that has seen a lot of success. Like the cursor founders dropped out of MIT and uh, you know, the Mercour team, you know, similarly super young team and founders that just were incredible builders and operators and executors. Um, and then yeah, I mean I think it's you know a lot of the like the familiar names and usual suspects, um, that just have proven to be uh, amazing technical um, founder factories. Kind of like Palantir is an obvious one but you know they really do have a culture of training people to like in the founder mindset and as a result they produce a lot of great founders. Ramp is another example stripe you mentioned. Um, and it's kind of interesting, there's like this rising generation of still young but fast growing kind of growth stage AI application layer companies that I think may start to become like these factories or breeding grounds for the next generation of startups. Companies like Sierra, uh, or Decagon or Harvey or Hebia Rogo, you know, company people that have seen success up close firsthand. Um, but again I think that like great, great talent can come from anywhere. And uh, it's certainly not limited to those kind of like big names that people think of for sure.

Speaker A: Um, I wonder maybe we could start with like. I really agree with you all. Like uh, there's like a few different profiles. Whether it's like you know, people coming out of uh, big research labs and or you know uh, I guess like the younger dropout from like tier one schools. Um, I guess like when it comes to like investing strategies since you guys cover every aspect of AI whether is um, you know, I guess like Word Lab is you know obviously Word Labs so or like you know uh, there is uh, different like for example I'll said that's kind of like a more sector specific AI. And then there is um, you know, Crusoe, which is like kind of like infrastructure area and then there is like different subsegment of like AI. How do you think about the investing thesis and uh, to evaluate different type of AI companies like uh, do you feel like we are just basically no one is actually investing in like foundational model anymore. But like people are really diving into the sector specific AI era and you know, where do you see the kind of like general trends go and then we can dive into some specific stuff.

Speaker B: Yeah, yeah. So the way we think about the universe of AI startups is as a technology stack with multiple layers and we, we invest up and down that technology stack. So we're full stack investors, we invest actively at each layer of that stack. Jensen has, Jensen Huang, the Nvidia CEO has recently been starting to talk about uh, AI as a five layer cake which is like closely mirrors the way we think about it. I feel like we were thinking about it as layers in a stack before the five layer cake analogy, but now everyone thinks of it as a five layer cake. But anyway, I mean I think it's basically right that the way we think about it, ah, at the bottom most layer is kind of like the energy and the physical infrastructure that powers AI, whether that's energy generation or data centers, uh, equipment and technology for data centers, the energy grid, et cetera. Um, and then a layer above that we think of as the silicon, the chips, um, the hardware to power AI. And obviously Nvidia is the dominant player at that layer. But there's a lot of interesting opportunities and new challengers and alternative approaches on the silicon side that we think are really exciting. Uh, a layer above that I would think of as like the transition from the world of atoms to the world of bits or going from hardware to software. And I think the kind of the foundational software layer is the models, the foundation models and of course a ton of probably the most famous well known AI, uh, companies are at this layer like OpenAI and Anthropic. Uh, but I think there's a ton of activity at the model layer beyond language. And so to your question around like are, are firms starting like stopping to invest in model layer companies? I don't think so at all. I think there's still a ton of opportunity at the model layer. But uh, one key trend is just modalities beyond text. So there's a lot of interesting frontier models being built in video, there's really interesting frontier models being built in robotics. You mentioned Generalist, which was an investment. We just led, um, a lot of interesting model companies being built in biology, material science, music, tabular data, etc. Like we're kind of discovering more and more data modalities as we go. Um, which I think is exciting. Um, anyway, so that's m, the model layer, a layer above that I think of as kind of like the tooling or infrastructure layer, the layer that enables you to take the core intelligence being built by these models and actually like productize them, commercialize them, operationalize them. So that's things like vector databases or RAG solutions or reinforcement learning as a service or these sorts of uh, different uh, software infrastructure. Then the top layer obviously is the application layer. Uh, I think long term my expectation is that's the biggest layer. That's where the most value accrues because that's where the rubber hits the road at the end of the day, uh, in terms of the technology actually being deployed in the world and making an impact on people's lives. Uh, and so we at Radical, we invest very actively at every layer of that stack. And it's not easy to do. Uh, and it requires a dedicated team that's like completely 100% all in on AI and just focused on investing uh, in every facet of the AI universe that's relevant. And so starting all the way at the bottom. You mentioned we're investors in Crusoe. We incubated this company Emerald AI, which is building a software orchestration platform at the nexus of AI data centers in the energy grid. Um, but we spent a lot of time kind of at that bottommost energy layer. Um, we are very active at the chips layer. We recently invested in Etched, which is uh, a new inference focused chip that's challenging Nvidia on the inference side of things. But one of our first investments actually as a firm was in Xanadu, a quantum computing company, um, many, many years ago, which just recently went public. So I think there's a lot of exciting activity at the, you know, in terms of alternative approaches to computation. Uh, we do a ton at the model layer. We always have, you know, we, I mentioned, we incubated Cohere, we incubated World Labs, we incubated Reka. Ah, um, you know, we, we just invested in Generalist, we've invested in companies like Periodic, uh, and Orbital Materials that are doing material science models. We've invested in companies like Leighton Labs and Nabla Bio that are doing biology models. Um, so we, we are, we continue to be very active at the model layer. Um, and, but I would say we're not, um, you know, we're not exclusively focused on research first or tech, um, you know, deep tech companies. We also do invest actively at the application layer. And you mentioned Outset, which is one great example. We're early investors in Hebia as another example. Um, and you know, plenty of other companies on here are kind of more like enterprise software application companies as well. So hopefully that provides a good overview for sure.

Speaker A: Basically it's like every segment and AI. Uh, I want to start with like you know you mentioned about you guys incubator like Coherer and uh, Reka and like other companies. Maybe we could start with like um, how does the incubation work? And um, how does like Radical kind of like um, think about the strategy since like the early stage round is like 650 uh, million. And then like what does a portfolio construction look like compared to maybe the later stage fund?

Speaker B: Yeah, totally. So I think that in terms of how we approach incubation, I think it goes back to the earlier point that like people matter more than anything and people are like the, the most important ingredient of any startup succeeding or failing. And so our incubation efforts are always centered around people first and foremost and kind of go from there. And so in the case of Cohere, like I mentioned, we're very close with Jeff Hinton and have been for a long time. And uh, Jeff like the uh, one of the co founders of Cohere, Nick Frost was Jeff student at the University of Toronto and another co founder Ian Gomez had also worked with Jeff. And so Jeff basically said to the Radical team like these guys are awesome. They're thinking about starting a company. They're amazing researchers, you should spend time with them. And based on that recommendation we started working with them. Um, this was like 2019, 2020, you know, pre GPT3, you know, a couple of years before ChatGPT, before large language models were like a well known concept and were consensus. But you know we got a lot of conviction in the size of the market, the opportunity they were taking. You know they were coming out of Google, even had been one of the authors of the uh, original paper. Yeah. Uh, so we had a lot of conviction that they were like top tier founders to back um, and so we spent months with them basically co founding the company, um, you know, helping uh, them think about hiring and product roadmap and technology plan and so forth. And you know we continue to be big investors. You know, we're still the largest shareholders, we're still on the board and uh, that's been a great journey. Um, but it played, I would say it played out in fundamentally similar ways with other incubations like World Labs. Uh, we've had a close relationship with Fei Fei for years. As I mentioned, she's a partner at Radical, she's an investor in the fund. Um, she's a dear friend and uh, close ally of ours. And so we knew that we were excited to be involved with her no matter what she ended up deciding to build. And so similarly with World Labs, we spent many months with her, uh, years even kind of helping ideate the idea and the plan and the vision and the concept. And then we're the founding investors in the company alongside Andreessen and continue uh, to be really close with her. And uh, I would say it's a similar story like Emerald. Just to take one more example, Varun, uh, the CEO is someone that I've known for 20 plus years at this point. We went to college together at Stanford and he's always been someone that I've thought incredibly highly of and is just, you know, one of the most successful and impressive people I've ever known. And so when the opportunity came up for us to do something together, I first and foremost had a lot of conviction that if Varun is doing the company, it's going to be a massively successful company no matter what. Now let's figure out what makes the most sense and the product and what market to go after and so forth. And in that case it was a really natural fit because Varun is one of the world's leading experts in energy markets broadly. Um, and so he really brought the energy side of things, we brought the AI side of things. And there's this moment in time where there's this really massive opportunity to try to figure out how data centers and the energy grid can fit together. It's kind of this huge unsolved problem. And so that was another example where it started with the person more than anything and then there's just profound founder market fit. Um, so I would say that's how we think about things on the incubation side. Um, but incubations are not the only thing we do. Out of the early stage fund we also invest in pre seed rounds, seed round series A rounds. Um, generally we lead rounds, um, as that's kind of our default approach. Although you know, we will sometimes co lead rounds, we will sometimes join rounds led by others. Um, and uh, and then similarly out of the, out of the growth fund, you know we m, uh most often are leading or co leading but also have some flexibility and and uh, of course the way we evaluate companies evolves depending on the stage. Um, and yeah, happy to get into that in more detail.

Speaker A: Yeah, for sure. I feel like because of all the AI companies right now are raising not all of them, but like many of them, especially from the Fundamental level like a lot of people are raising mega rounds and then how do you think about like uh, the valuation of the company as well as since many of the researcher um, led companies are uh, very I guess like research raising a lot of money right now. And then where do people typically think about the spending of the company? Since nowadays token maxing are getting a lot of attention because of whether it's big tech company or um, just people are not necessarily spending um, the token in the most effective way. How do you think about I guess for the CEO or the founder to um, allocate the money into whether it's like developing the product to go to market and everything else.

Speaker B: Yeah, I think the concept of token maxing is a transient phenomenon. I think we'll look back on it in a few years and think it was like a really silly, uh, somewhat nonsensical concept. Honestly from first principles is obviously not a good measure of anything worthwhile just to like the sheer, the sheer amount you can spend. Um, but I also think it's like in this transitional phase where companies are just starting to understand and experiment with and master like how to get the most out of these, out of these brand new technologies that are so powerful. Like it's understandable why it became like a stand in proxy that's better than nothing and is like a course uh, indication of like how much is my team adopting AI. But it's like so obviously so ripe for good hearts law issues where like if you, if you're, if you're optimizing for that, you can obviously just max that out in a way that's not helpful or productive. So and I uh, think that tide is very much turning at this already. Like there was, there is this window where there was just so much pressure and so much focus on like we need to be AI first, we need to be AI native. We need to be using these tools as much as possible. So like everyone go tokamax and I think that people are already starting, there's very clearly a transition where people appreciate okay, we need to get more nuanced than that in our thinking because budgeting matters and costs matter and we ultimately want to have unit economics that look good and we don't want to just spend infinitely. Um, and so I think there's a huge set of opportunities that's companies that are being built today and that will be built around optimizing companies use of intelligence and using the right models for the right tasks. And you don't need to use Fable to answer a question. About uh, some basic question about what's the capital of France or something. Um, and many everyday tasks in the enterprise don't require the frontier models which are way more expensive than using a much smaller model or an open source model. Um, so I think this general idea of model routing and understanding which models you can use, whether they're in house models vs closed models, behind an API vs cheaper lagging previous generation models, I think there's a massive set of companies and products to be built there And I think OpenRouter is one interesting early leader in that space. Um, and then just more broadly I think getting, focusing on optimizing um, token use and getting the most intelligence per flop or intelligence per watts. I think that will be a huge focus going forward. Um, and there's just so much low hanging fruit today. The models are amazing but they're used so inefficiently. And the intelligence per watt today I think we're going to look back on in a few years and it's going to be crazy like how resource inefficient these models are being used. So certainly I think that's a big. Yeah, that's a big meta trend to keep an eye on going forward.

Speaker A: M. You uh, also wrote a lot of different articles in the AI space. I want to start with since I think it was 2025, you did 10 AI prediction for 2026. Um, do you have a new version of this? I really like what you said. Therapy will go pop. Like oh man, I will not. Because that's like, um, you know, that's like kind of like not very obvious a year ago. Maybe we could start with like, you know, how did you um. Yeah, is there like any trends that kind of like trend and why do you feel like, you know, back in time, how do you kind of like see this before everyone else?

Speaker B: Yeah, yeah. So I've been writing these annual prediction articles every year since 2020. Uh, in December I write a list of like 10 predictions about AI for the coming year. Uh, and then the following December I grade them publicly, which I think is really important. Just M in terms of like having some, you know, holding yourself accountable, having some intellectual honesty around it. So uh, and so honestly I like, I enjoy doing the grading even more than the actually making the predictions just to like. Because you learn so much by looking back and seeing, okay, what did I get right? What did I get wrong? Um, I also like, I would say a couple important principles when I'm making these predictions. One is like, I try to make them really Specific and like falsifiable or verifiable. So it's not like agents will become important this year. You know, like it's like very specific like OpenAI will go public. Anthropic will or sorry anthropic will go public opening. I will not obviously that's like it will be clear if that's true or false at the end of the year. Um, and then I also try to pick things that are non obvious and provocative. Ah and honestly I will share that I sometimes make predictions that I don't actually think are super likely to happen but are interesting to think about as possibilities and then get people thinking and stimulate thought and discussion and controversy. The 10 that I've made for 20, uh, 26. It's interesting to yeah. Do a kind of mid year review of them. I think some of them I'm feeling good about some of them I think we're totally wrong. Like this one, this one that you have on the screen number four is, is. I think this is totally, I got this one totally wrong. There's, there's like just been more and more focus on AGI and super intelligence. Uh, but uh, you know we'll see what happens the next half of the year. Um, but other ones like the. Yeah, there's a bunch of them that um, I think will be interesting to watch play out. Um, and then I also, I publish. It's good timing that you brought it up. Just this past Sunday like a few days ago I published a new set of predictions for five predictions for 2030. So like looking forward not just one year in the future but five years in the future which is a very fun exercise. And uh, I tried to make those, those ones I think are particularly spicy and provocative and I've gotten a lot of like interesting and fun feedback. Obviously some people agree, some people disagree with different predictions but I think I just find that they're great conversation starters and great like thought uh, exercises to think about, you know, in what, in what non obvious ways is the world going to change? And I, I guess one last thing I would say is like especially for the ones that are five years in the future, like if they don't sound a little ridiculous, they're, they're not good predictions because the world is going to change so much that like you need to be thinking in ways that feel a little uncomfortable today. Like this, this one is a good example like the TSMC ASML1 today there's such prevailing wisdom that TSMC and ASML are these like unassailable leaders. Um, but the prediction that I made is that, like, their monopolies will actually be broken in the next five years. And you can read the article to see. See my justification of that. But anyway, it's a fun. It's a fun intellectual exercise.

Speaker A: M. Um, can we talk about, like, how do you actually come up with a prediction? Because I feel like when I was like, browsing through it, like, I feel like it was very detail oriented. So, for example, you know, uh, let's say like, Anthropic, hire the CFO of Airbnb and like, from, like the, you know, human level to everything else. Like, I guess, like, what would you say it's like the different strategy that, like Anthropic and Open, I had that kind of like, set them apart because I think, like, Anthropic is kind of like, they pick the lane of like, being focusing on, like, code, um, generation or like, focusing on, like, the developer, um, as a star. And then nowadays, like, and then obviously, like, you know, OpenAI and like, everybody else are like, focusing on, like, let's do consumer. Let's do, you know, uh, AI generated videos and like, you know, split, Um, I mean, I want to say spread themselves too thin, but more focus on like, the consumer playing than like, enterprise. Um, but obviously, like, I feel like Anthropic did like, um, such a good job on like, monetizing everything. So maybe we could start with like, you know, in your observation, like, what are.

Speaker B: What.

Speaker A: What do you think, like, Anthropic did right, versus, like, everybody else?

Speaker B: Yeah, I think you summed it up well. Like, I do think what has proven to be the case is one of the Anthropic's great strengths is how maniacally focused it's been. And they. And like, Maniac. Maniacally focused. And also, like, they made the right bets. But, like, obviously both of the, like, it's important to be focused, but it's also important to be focused on the right thing. Um, and as you said from the very beginning, from the company's origins, they had a lot of conviction that coding was this really important force multiplier. And if you could build AIs that were really good at coding, that would be a foundation that would m. Enable you to have an AI platform that was good at so many other things and would be a leverage point to improve your own AI systems more quickly. Um, and that bet paid off massively. It enabled them to build models that are the best of code and set the stage for the explosive success of Claude Code and the incredible revenue growth that they've been seeing over the past couple years. And in hindsight maybe it's obvious that today the dominant, um, driver of AI commercialization and revenue and use case is coding. That's the number one killer app. Uh, and so Anthropic got that really right. Uh, and by comparison, as you alluded to, OpenAI has had way more sprawling ambitions. Ah. And they've been kind of all over the place. And there's a period a couple years ago where it was like a sign of strength. It was like, oh, OpenAI is going to conquer the world. They're doing enterprise AI, they're also doing consumer AI, they're also doing consumer hardware with Jony I've. And they're also doing robotics and they're also designing their own chip. And Sam, uh, Altman co founded a brain computer interface company and they were rumors for a while. That same album was like thinking about buying a space, a launch con letter. It was like, oh, they're, you know, what an amazing company. Uh, but it, I think, I mean, I think the phrase you used is accurate that they did spread themselves too thin. Um, and I think that they've come to realize that this year and there's like been a pretty conscious narrowing of focus and you know, OpenAI leaders have spoken about this publicly and they shut down SORA and they deprioritize their AI for science efforts and I think they are trying to refocus. Um, and uh, you know, and like, you know, I think to their credit Codex has made massive strides over the past few months and I like, it's getting a lot of developer love and like OpenAI, uh, obviously still has so many great things going for it and so many advantages. It has an amazing team, you know, incredible fundraising machine, a ton of capital, a ton of compute, amazing models. Like they're, they're by no means, like it's by no means settled that Anthropic has won forever. But I do think that like that focus proved to work really well for Anthropic. But you know, I think both companies are still on amazing trajectories. Both companies will be public companies before long. Um, and you know, I think there will be many more chapters in their competition.

Speaker A: Um, what would you say are like the things that you kind of like saw before everyone else? Because like um, like you mentioned like, you know, that let's say like Anthropic hire the CFO or like, you know, from the talent perspective, like, and then we briefly touch upon like you Know, you mentioned that like, you know, people. The Frontier Lab, the, the Frontier Labs. Like only I guess like people don't leave Anthropic, but people leave everywhere else. Like why do you think that is?

Speaker B: It's, it's amazing. I mean it's, it's incredible.

Speaker A: The opposite. It's like people are joining them. Like the functions are Dr. Them. It was like crazy.

Speaker B: It's, it's unbelievable. I mean like, you know, I can only speak from the outside, like not, not being inside and experiencing it firsthand. But it seems, it seems like the company just has an incredible culture, incredible followership, uh, incredible like alignment and sort of like personal alignment with the personal passion and personal beliefs of, of the employees. Um, but yeah, it's, it's like incredible retention. You know, very, very few people leave. And then to your point, like especially in the past few months, like this trend has just accelerated of like everyone wants to join. Like everyone in the world wants to join Anthropic. Like Andrej Karpathy joined. Like that was obviously huge news and he had held out for years, you know, rejoining a Frontier Lab. And everyone was like, what? You know, when is Carpathia going to join a Frontier Lab? And he didn't want to. But then, you know, the fact that he finally joined Anthropic is very telling. I mean, John Jumper joining from D Mind is crazy like that. You know, he won the Nobel Prize with Demis. You know, he did alphafold with Demis and was such a pillar of the DeepMind organization. And for him to leave and join Anthropic, like that's huge. And you know, I'm m sure folks have probably seen some of the like, like the, the lists of like CTOs of public companies, like the CTO of Workday recently left to join Anthropic as a, as an individual contributor. The CTO or the former CTO of Box like this, the like you know, the online storage company left to join Anthropic as an individual contributor. So yeah, it's just this like insane gravitational force and talent magnet of people that are joining and you know, to our discussion at the very beginning, like the people matter more than anything. And so I do think that's like an incredible, an incredible advantage that they have for sure.

Speaker A: Um, I wonder. So like one of the things like um, we can uh, I guess I am like curious about your thought is like from maybe we could use like any of your portfolio coming from like sourcing, diligencing, winning to exiting. Like what's your kind of like thought process on these from maybe like, from sourcing perspective, since like we chat about like people, whether it's like leaving big labs or like you guys incubate them, or like, you guys kind of like start the conversation extremely early, like years early. And like, how do you think about, from the sourcing perspective on um, actually getting the best researchers? Obviously, like, nowadays, like, uh, when you think about researchers, you think about like people hosting these like researcher poker nights or like research paper, paper reading that it's like, but are they the best researchers? And since like, you know, everyone every year is going to these like mega conferences like neuralips and like neurips and then like, um, you know, obviously there's like people's name on the research paper, but who is the best researcher as like an entrepreneur? Um, in general.

Speaker B: Yeah. So I think, um, I think like the most important thing when it comes to sourcing and building relationships with future founders is like, you know, the, the, the, the more long term the relationship can be, the better. And so like, we really make an effort to not like, meet a founder when they're starting a company and raising around, but like, meet them years, you know, ideally years before and build a relationship with them when they're, you know, when they're inside GDM or when they're inside OpenAI or whatever it is. And um, you know, and just like, build a genuine authentic relationship where like, it's, you know, you enjoy chatting and sharing thoughts and you can be helpful for one another and uh, you know, maybe they will end up leaving and starting something and maybe not. But it's just, it's so much better to get to know people ahead of time from both directions. Like, the potential founder can get a real sense of like, is this actually an investor that I would want to work closely with and you know, potentially have on my board and to partner with on this journey. And from, from the VC perspective, you know, we can get a much clearer sense of like, is this really, like, beyond the fancy title of like, okay, they're at OpenAI or whatever. Is this really someone that I, that I have a lot of conviction is going to be an exceptional entrepreneur and like, it uh, has all the characteristics that are required for that. Um, and so that kind of gets to your question of like, how do you, like, which researchers are the best founders, basically? Uh, and that's like, it's a question we spend a lot of time thinking about and talking about about because we back, you know, we, we have always kind of oriented toward backing research Founders, technically oriented founders. And you know, we've had a lot of success doing that. Um, but it's not straightforward. And there are so many examples of like famous researchers or like very accomplished academics who, you know, did not make good founders at all. Um, and so it's definitely not as simple as like, oh, this person is like a famous, well known researcher, therefore they are, it makes sense to invest in them as a, as a startup founder. And like I think the most important and obvious dichotomy is like, do they, you know, is it someone that just wants to do science, that just wants to do experiments? Or is it someone that's genuinely commercially oriented, uh, you know, has high eq, is able to think strategically about, is a talent magnet in terms of like hiring people to build an organization, is going to be able to run a large organization, is able to think of whether or not the experience matters less. It's okay if they don't have any previous experience in a commercial role, but do they have the commercial instincts and the commercial orientation to think about what is a business model here? Who are our customers? What do our customers want? What value can we offer to our customers? How do we think about, how should we think about partnering with them and so forth? Um, and none of those things are straightforward to suss out. I think it really just involves spending a lot of time with founders and prospective founders and getting a feel for that and talking with them about commercialization and productization. And how would you think about this and how would you think about that? Um, and yeah, I mean there are founders that we've backed who had amazing research pedigrees and then they turned out to not be great entrepreneurs. And there were founders that we, their founders that we backed who had amazing research pedigrees who turned out to also be amazing entrepreneurs. And so it's uh, yeah, they're, they're related but distinct skill sets.

Speaker A: What do you think about the diligencing process? Since like, I mean like many of them are like inventing things on their own. Like uh, whether it's like um, like actual foundational model or you know, the uh, product itself. Like um, I guess like um, how long do you think it take? So since we mentioned about like you know, some of the researcher, you may know them for years. But what about like, you know, I've seen some of the newer, like bigger rounds are raised by let's say like a few entrepreneur that, who have done something that's like really critical and then they all like kind of like merge together. To like create a new company. Um, but like do you think those are like, like how you really diligence these kind of like companies because of it's kind of like people are raising just based off like their name in the research industry or like kind of like their track record of being able to done something. But like I guess like how long do you think the diligence process will take? And then do you feel comfortable to really see some sort of like product and traction? Like let's call it like I guess like design partner that's signed to like certain things. Like how do you feel comfortable about like investing in like some mega rounds?

Speaker B: Yeah, yeah. I mean I think in terms of the how to diligence people, like I think you know, the most important thing is spending time with them. Um, I also think references are super important. Talking to people that have worked with them extensively in the past and getting candid perspectives from them. I think that's one big reason why like we as a firm being so deeply enmeshed in the AI research ecosystem and AI networks matters a lot and it's such a big advantage because we can, you know, we can call up and talk to you know, the most important senior leaders at these different organizations and get candid feedback on like oh, you know, this person, this person was incredible and this person like really was the key driver of this thing versus like oh, this person actually was kind of toxic to work with or difficult to work with or like didn't contribute as much as, you know, they act like they did that sort of thing. So you know, of course the kind of referencing people extensively is important. Um, but uh, to your question around, I think this trend of neolabs and massive rounds, I think um, early stage venture has always been about backing amazing people with very little that's actually been proven or built yet, but believing in the person, believing in the vision, believing in the market. And so I think that is something you just have to be comfortable with as an early stage investor is investing very early. Before there's obvious proof points, but having conviction in the team. I think what's the wrinkle with this current wave of neolabs is the valuations that these companies are raising at uh, prices in so much future success in such a way that the risk reward balance in a lot of cases just is very fundamentally off. It's one thing to say I believe this is a big swing, it's going to be hard, it's going to be an uphill battle, but there's a big upside and this team is great, so I want to invest in them. Um, but you know, if you're doing that at 50 million or 20 million post, it's a lot different than if you're doing that at 2 billion posts. Um, you know, to invest at a company at 2 billion post and then have the return that you need on that to justify it. Like you know that would have to be one of the, you know, one of the great venture backed outcomes of all time. Uh, like there's very few companies, very few venture backed companies that exit at like $20 billion or more, you know, much less $100 billion. So yeah, in a lot of cases like I do think the valuations have gotten crazy. The risk reward balance is very out of whack. Um, but it's intuitive to understand why because Even if there's 100 company 100 Neo Labs that are founded and that raise money at 1 or 2 billion dollars in their first round, um, and 99 of them fail, but one of them is the next anthropic, one of them can follow the path that anthropic followed and went from zero to a trillion dollars in five years. Um, that makes it worth it. And obviously investors try their best to have some discernment so that they can position themselves well to be picking the one out of the 100 that is anthropic. But I think just seeing the magnitude of growth that these companies have had anthropic and opening eye and others, uh, and the increase in value in such compressed periods of time is unprecedented in history. Like no, you know no previous technology wave saw this rapid growth this quickly. And so I think that's that, that's the like economic justification for why these, you know, these crazy rounds are happening.

Speaker A: Uh, what about how do you think about like in terms of like uh one thing is like obviously like picking a winning are kind of like very ah, jointly together. Because I feel like nowadays like uh, because of like let's say um, it's very similar to what you said about like you know some of the company are the valuation are like too high because of um, you know it's hard to justify as a fund in terms of like the return but I guess like when it comes to uh, you know, winning the deal that's like everyone wants whether it's like Word labs or you know, some of these other coming in your portfolio. I'm sure it's like chased by like tons of people. And how do you think about the winning part as a fund to kind of like Help them succeed later or, you know, I'm sure you guys probably know many of them like very early on, but like, but nowadays since everybody is investing in AI, all the mega funds like, are um, very emphasizing on AI. So like, I guess, like, um, how do you kind of like, I would, I wouldn't say compete, but like kind of like work with the current AI climate.

Speaker B: Yeah, I think, you know, again, I think so much of it uh, comes down again to personal relationships and like having a really genuine, strong, positive bond with the founders and have, you know, I think founders appreciate like who I pick as my partner and my investor in the first round and the second round. Like, those are really important decisions. Like, it's kind of, you know, the analogy is somewhat overused, I guess, but it is in a lot of ways like a marriage, a professional marriage. And like you're gonna be teaming up with this person for years or you know, a decade or more. Uh, and so I think founders appreciate that that matters a lot. And so just being able to establish that strong personal bond and a level of trust and a sense that like, this is someone that I want to have as a sounding board and like, it's not someone who's just going to be like a blind cheerleader that just like supports me and is like blindly positive all the time, but also is not like, you, uh, know, founders also don't want someone that's like going to be criticizing and writing them and difficult to deal with all the time. And so I think like, and different founders, for different founders, there's different VCs that are the best fit. Obviously people have different personality styles and types and so forth. But I think that at the end of the day that personal human element matters so much. Um, and then of course there's a lot of other dimensions to it. And I think what makes Radical special is we are a VC firm that has been built from day one, purpose built to support AI companies and uh, everything that's unique about an AI company. And we've built our entire team around that. So we have a whole talent team whose entire focus is helping our companies hire AI talent, find and identify the very best AI people and hire them. And to the point around how scarce AI talent is, that's an incredible, uh, value add for companies that no other firm in the world has. We have a huge in house talent team that's 100% focused on AI talent. That's like, you know, if you talk to any of our portfolio founders, like it's, that's such a Valuable resource for them to have just this additional engine of talents. Um, you know, we have a whole internal compute team. You know, if there's, if there are two things that AI founders worry about and spend their, like lose sleep over it's talents and computes. Um, and so we, you know, we have a whole team that helps our founders and our portfolio companies like get their hands on the right compute, negotiate deals, get preferential terms, get access to the latest chips before anyone else does. Uh, and so you know, those types of offerings make a big difference. Um, and I think, I also think it's just like the, the um, community of founders that you have. And I think people, founders appreciate that like partnering with Radical and having Radical as an investor kind of welcomes you into this community of leading AI founders and AI startups and top AI researchers across different domains and different spaces. And so I think it kind of becomes like a positively reinforcing cycle of founders want to be a part of the Radical network and to be connected with this ecosystem of other amazing AI founders.

Speaker A: Uh, I totally feel you about essentially because of Radical already invested all these top AI companies. So it gave you guys fair advantage to really land the next generation of AI companies as well as the infrastructure about whether it's talent or access to um, chips, uh, the newest technology, um, I wonder. So I want to maybe quickly touch upon two small subsector which is one is physical AI. You um, guys, um, invest in um, generalists. And how do you think about the physical AI space right now? What is actually investable since there are so many robotics companies these days and coming out of whether it's like Nvidia or they're selling to the government and there's many subsector. Is there any particular sector that you personally feel really bullish about? Whether it's like consumer or um, industrial robots?

Speaker B: Yeah, I think robotics is an exciting space. Uh, building AI that works in the real world is harder than building AI that works in the digital world. And so as a result of that, the kind of the, the level of progress in robotics inevitably lags a few years behind the progress in large language models or agents, you know, digital agents. Um, but I think we are nearing this inflection point where these models are really starting to work in a profound way. Um, and you know, uh, in less than. In the, less than the past year, like maybe the past nine months, we really have crossed this threshold where these General Purpose Robotics foundation models are starting to have levels of performance and also levels of reliability where like it's viable to deploy them commercially in a wide range of different use cases. So I am very bullish on this notion of being able to build foundation models that are general purpose that, you know, that can do few shot learning, that, you know, where you don't have to build one model to do one robot task and a different model to do a different robot task. But you know, very similar to the like GPT three moments, uh, where you realize that you can build one model that can do all sorts of different things, um, and you can deploy it across a wide range of different form factors. I think that, I think we're getting pretty close to that moment in robotics. You know, I would say we're not quite at the like GPT3 level, much less the ChatGPT moments, but we're getting close. Um, and so that's why, you know, we were super excited to invest in Generalist, you know, similar to what we were saying about talent in the world of AI broadly. Like, there's just so few top notch world class teams and individuals who know how to build AI robotic systems at the true frontier. And Generalist is one of those very few teams. Just has an absolutely stacked group of folks. And I would, similar to our conversation about anthropic, I think Generalist is maniacally focused, uh, and has a very particular vision of what's going to work and has been executing on that and it's, it's becoming clearer and clear that it really is working. So, um, you know, we're very excited about, and of course there are other, you know, other exciting companies, uh, in uh, the robotics space. But I think that that approach is one that we're excited about.

Speaker A: Uh, I want to uh, maybe like, I want to be mindful of time. Maybe we do a one real question and then 30 second firearm for you.

Speaker B: Okay. I do, uh, I do have a hard. Yeah.

Speaker A: Okay. Who I started. Okay. Uh, what's your favorite book?

Speaker B: Girdle, Escher, Bach.

Speaker A: Uh, who would you invite to your dinner party?

Speaker B: Interesting. Um, uh, can they be dead or do they have to be living?

Speaker A: It's your party.

Speaker B: Okay. Alan Turing.

Speaker A: Uh, who made the biggest impact in your career?

Speaker B: Um, I had a professor at Stanford, actually professor of philosophy who's very, uh, influential for me.

Speaker A: Uh, where can we find you outside of work?

Speaker B: I love traveling as much as possible. I just, just got back from Switzerland and Spain and you know, sometimes for work, sometimes for, for pleasure with the family. But yeah, I love, I love traveling abroad as much as I can love it.

Speaker A: Well Rob, thank you so much for coming on the show today. It was such an informative conversation about everything.

Speaker B: AI, thanks for having me. It was a lot of fun.

Speaker A: Uh, let me.