[0:00:00] At the end of his talk, he said something like, and things are never going to be chill again. They'll never be. And like now we have these like cute sweatshirts and it says things will never be chill again. Welcome to Frameworks for Growth. I'm Christina Cassioppo, co-founder and CEO of Vanta. And I'm here with Danielle Amode, president and co-founder of Anthropic. So prior to starting Anthropic in 2020, you were VP of safety and policy at OpenAI, led policy teams, engineering teams.
[0:00:35] Also early employee at Stripe and led risk and recruiting teams there and started your career in international development. Thanks so much for being here today. Thank you so much for having me. So Anthropic describes itself as an AI safety and research company that's kind of different purposely than a kind of standard technology company. What does that kind of mean to you all in practice?
[0:00:56] So first of all, so great to be here today. Thank you for taking the time to chat with me. So I think we really focus on how to develop our tools and systems in a way that is robustly safe really across our stack. And what I mean by that is we apply
[0:01:14] Technical training techniques to how we actually train our models to ensure that they are aligned with human values. We use a technique called constitutional AI that helps us to imbue Claude with a sense of ethics. There's a number of different documents that actually went into helping to train Claude with constitutional AI. It includes things like the UN Declaration of Human Rights.
[0:01:38] How we ensure that the technology is safe when we actually release a new model or a new set of products, we have a framework called the Responsible Scaling Policy. And our Responsible Scaling Policy is essentially a set of guidelines that indicate how and when we will release models to ensure that they're safe.
[0:01:58] We use different level of AI safety levels. There's five of them, and they're actually modeled along the lines of biological safety levels for people that do research with biological applications. And really what we look for is defining a concrete set of safety and security requirements at every ASL level before we release a model. That's very neat.
[0:02:22] And you're kind of touching on some of this, but Anthropic is a research lab and doing a bunch of kind of deep research and also trying to commercialize the technology. Most startups aren't trying to both do a bunch of very real technical research and commercialize at the same time. How do you balance that tension? I think this is one of the most interesting and unusual things about running a company like Anthropic is we're almost, you know, part.
[0:02:50] Research Lab, you know, part tech startup. And we have a huge emphasis on our public benefit mission as well. And really thinking about how to balance all three of those things in service of the mission and in providing value for our customers, I think is one of the great joys and excitements of getting to work at a place like Anthropic.
[0:03:08] In terms of the research to product balance, really what we found is that having this kind of collaboration between product and research really helps to make both of those teams much stronger. And in particular, so much of what we see is that as new research capabilities are developed, it actually opens up new avenues in the product that wouldn't have existed before the technology itself advanced. A great recent example of this is
[0:03:36] Claude has been particularly useful for developers in writing code. And many of the products that we've developed over the course of the past two to three months have really been geared around making that process easier for software developers, whether they're at a big enterprise or a mid-market company or at a budding startup. And if we hadn't known that the models were going to be really good at coding, we wouldn't have necessarily designed.
[0:04:00] Parts of our product that way. And so I think in some ways there's this iterative process where as we're getting more customers using us for coding, we're also helping to train the models to make them stronger in coding and then providing more product support in that area. Makes a lot of sense. And we use a lot of that at Vanta too. That's great. Glad to hear. Happy customer. Yeah, very much so.
[0:04:21] Actually, curious to talk about public benefit corporations. Most startup founders do not decide to be public benefit corporations, especially on incorporation. Should they? You know, when we were starting Anthropic, we actually spent a reasonable amount of time thinking about what kind of structure we wanted to incorporate as. And we had a little bit of a benefit of having seen sort of different models up close. Some of what we were thinking when we incorporated as a PBC was,
[0:04:51] What was a very practical way to balance both the commercial interests that we foresaw we would have with the public benefit intentions behind founding a company like Anthropic that was so invested in topics like safety?
[0:05:07] And what we ultimately decided was that public benefit corporations are a really nice blend of those two areas. It's a C corporation, so it is structured in most ways, you know, very similarly to a Delaware C corporation. But the one exception is that we have the ability to balance commercial needs against public benefit.
[0:05:26] Well, between OpenAI and Anthropic, you've helped build two of the kind of iconic AI companies of our generation.
[0:05:56] I'm curious, when you think about literally building companies for people and literally building these companies and hiring them, any hiring lessons or maybe things you learned at OpenAI that you've applied at Anthropic? This is probably not surprising as an early employee and founder, but I think so many founders and early employees talk about culture and how important culture is when you're growing a company, particularly one that's scaling really rapidly. And I just want to echo that I think the most...
[0:06:24] Valuable currency that you have when you're a small and quickly scaling company is the culture of that company. So to me, culture is really an expression of the company's values. And I think the more clearly it can feel really hard and very fuzzy to sort of pin down.
[0:06:40] What is your culture and what are your values? And we go through the process. We write the mission statement and we write the values and we have the culture interviews. And it never completely captures the feeling that you have when you walk in the doors and you're an employee and you're like, I know what it feels like to work here. But I think the effort to really invest in that.
[0:07:01] process early and say, okay, it can never really be quantified. But if I had to quantify it, how would I describe our culture, right? What are the core values that underpin who we are as an organization? And how do we make sure that people that come to the company are on board with that? And I think probably the biggest lesson I've learned at Anthropic is that you should really be as open and communicative about what your culture is as possible. Because the people that are excited about that
[0:07:29] Set of values and that vision are going to be attracted to you. They're going to say, wow, like we love that you are a public benefit corporation and you're balancing these two things and you care about this set of topics. And for people who that's not their cup of tea, it's much better to learn that early, right? You want them to read your website and say.
[0:07:48] Well, that sounds nice, but that's not really what I'm interested in. And so I think it's this sort of delicate balance of wanting to attract the right kind of people and really, really having this very extreme almost emphasis on ensuring cultural alignment in your early employees. And how do you talk about the kind of values in a way that is clear? Does that make sense? Because I think the critique of startup values is, oh, they often all kind of sound the same.
[0:08:15] What's the like anthropic spin on them where you can, you feel like you're just clear with people about what is important to you all? Yeah. Number one, I would say if your values are not
[0:08:26] Making people kind of like, if they're not creating tension, they're probably too bland, right? So it's like, be kind, be direct. Everyone wants that. And so if there's not an intentional tension to them, probably they're not specific enough. The second is that you need to be able to back up any value with 10 examples like off the top of your head, right? A value that we have.
[0:08:48] is do the simple thing that works. And this comes mostly from our research team saying, we're not actually here to have a bunch of fancy algorithms for no reason. We're really trying to make the models as useful as possible for people. And there's no extra points for sort of being complicated or writing these fancy documents or overthinking things. But I think that value applies outside of just research for us. And off the top of my head, I could explain how we use that not just in research, but in product, in engineering, in our office.
[0:09:17] Selection choice, right? Like in how we set up, you know, our offsites, right? And I think anytime a value is one where you can just point to examples of it really easily, it's more likely to be sticky because people will run into it and feel that value. Yeah. Okay, actually, so building off of that and for the founders watching. Yeah. When you're making an office choice or something in product development, how much do you tie that to your values? Like kind of how do you, you could rant off the example. Yeah. How do you get it so...
[0:09:46] The 200th person hired at Anthropic can also kind of pull up those examples.
[0:09:50] Yeah. Number one, I actually think writing them down and talking about them in public forums within the company is really valuable. So in our all hands, we try to mention a value, you know, every all hands. We don't always do it perfectly and you don't want to shoehorn it in. But I think really emphasizing and just making the language, you know, super comfortable. You have to make sure it actually resonates with people. Otherwise, it feels like sort of painful propaganda. But if it does resonate with people, it's actually something that they volunteer themselves. Right. And so I will hear other.
[0:10:19] employees in the company just say, oh, well, we were considering hosting this event off-site, but we just decided to host it at the office because we wanted to do the simple thing that works. And I think I always kind of light up when I hear a version of that or with a different value because it means this is actually something that's practical for people and it applies to them.
[0:10:39] And the other thing I would say is don't be afraid to retire values if they're not working for you. So sometimes something feels really right to you as the founder, but it's just not sticky. And it's maybe just you're saying it the wrong way. It's not necessarily that the underlying core part of it is wrong.
[0:10:54] But I think just feeling comfortable iterating until someone says, actually, that lands with me better. People should be memefying themselves. And another thing I think is interesting is some of our values, they didn't come from us, from the founders. They came from the company. Like, someone just started using a term, and then we were like, that's actually a value of the company. Like, that's great. And I think it helps employees feel like they have ownership over the values as well. It's not something being forced on them. I love the memefying of values. Any good stories there you can share?
[0:11:23] I mean, I think there's some really funny things that kind of came up in like our earlier days around swag, where like at an offsite that we had maybe two or three years ago, Dario was just talking about, you know, scaling laws and how quickly everything was going to grow in AI. And even those of us that like really, really believe that, I think, you know, now two years looking back, it's just so much crazier to see just like the rate of growth that's been happening in artificial intelligence and the adoption curve.
[0:11:50] And at the end of his talk or in the middle, he said something like, and things are never going to be chill again. They'll never be. And like now we have these like cute sweatshirts on like I have one on my toddler. You know, we have like the little kid version and it says things will never be chill again. We give them to new parents for like their babies. So I think it just like sometimes they just kind of like emerge. But like Daria wasn't intending for that to be like a meme. But then people thought it was so funny and it felt so accurate. And then there were all these like nice ways that we could sort of apply it within the business. I like that. Yeah.
[0:12:20] Any values you've retired over the course of the company? There are a few. I mean, move fast and safe things I love, but it feels a little too cheeky for us. I think we just like ended up adjusting that one because we were like, okay, that's just, it was like sticky sometimes, but not everyone loved it. And so we're like, if not everyone loves it, people are not going to use it. Outside of the values and outside of the like true platitudes, the be kind, like be direct, what's a non-obvious trait you look for when hiring? I think...
[0:12:49] One of the most important qualities, especially at Anthropic, is a willingness to talk about a wide range of topics respectfully. And so I think for us, you know, we work on a very, you know, cutting edge frontier technology. And a lot of the things that
[0:13:09] Kind of bump into it can feel complicated or controversial, right? So AI has incredible potential to help people. It also has incredible potential to be misused if we're not careful. And so how can we make sure we have to address those challenges at Anthropic? We view that as our core job. And so if people are not willing to have respectful disagreements and to be very open-minded, right? So even to say we're coming from completely different opinions about this topic, but...
[0:13:36] I have to be able to engage with you respectfully on those topics.
[0:13:58] It's not like we're looking for a particular belief system. We're looking for the ability to say, hey, I felt this was the correct thing, and maybe it wasn't popular, but I really stood by this. Or I changed my mind because I actually realized I wasn't right about this. Very neat. When you go through that interview, how do you prep people for that interview? Yeah, it's a great question. It might have changed since I gave it, which is now sort of a while ago. But we actually give a little bit of a preamble about why we're asking these questions. To say, hey, this is actually...
[0:14:27] We're really trying to show you what Anthropic is like and some of our values and why we've sort of chosen these questions is to kind of give you a little bit of insight too into like what it's like to work here. And for the vast majority of these questions, there's no right or wrong answer, right? It's not like, oh, you have to say this type of topic is, you know, was a place you held your ground. But we really want to understand your thought process and we want you to be honest, right? Like there's no right or wrong.
[0:14:54] Answer to the question. So like you don't have to prep for it. We just are trying to get to understand your sort of thinking process. Yeah.
[0:15:02] I ask because we have a Vantive version of that interview, so different setup, but it's the same thing. And we used to, in the early days, not prep people for it. And they would generally get quite confused by what was happening. So that's kind of why I ask. Again, you don't want to lead them. But it is kind of like, here's why. Yes. I think the context is so helpful. So they're not like, why are you asking me this question? Exactly. Imagine when you're...
[0:15:26] Recruiting, I mean, folks across Anthropic, engineers in particular, you want to balance the thoughtfulness and debate and keeping simple things simple. How do you kind of tease that out, maybe especially with engineers? So I think for both engineers and research, we are always in this really interesting dance of when are we investing in kind of the fundamental research?
[0:15:54] And I think this is very similar across a lot of companies. But I think at a place like Anthropic, it's particularly challenging because sometimes investing in fundamental research that doesn't necessarily look like it's going to have immediate returns for customers is the best way to create breakthroughs. And so I wish I had a perfect formula. It's much more of an art than a science.
[0:16:20] But at least on the research side, we really try to diversify the amount of time that we're spending optimizing particular qualities in the model or tools or systems and just some amount of open exploratory research time for our team. I think on the engineering side, it's similar. We want to give some scope for people to be able to really rethink and reimagine what products we're building and why. And I think it's highly related to the fact that
[0:16:49] Because the research is developing so quickly, kind of like in that coding example, you don't always know what's coming down the pike and things can change on a dime, right? And you can suddenly say, well, there's this incredible new use case and we should actually build a product around it. And so in some ways, we always want to be giving people the space to just reimagine how is it that we're even providing value to our customers?
[0:17:09] And we also have to set a product roadmap and build towards that roadmap and make sure that we're doing all of the very important but sometimes unsexy engineering work that needs to happen. And so I would say it just falls somewhere in the middle of the 60-40 or 40-60 or 50-50 with periods of time where we're really sprinting towards a particular release that we've been working on for a long time. But I do think a little bit of that open-ended exploratory mindset is an important quality for us.
[0:17:39] And then how do you kind of bake that into your processes or do you not, right? Like Google famously had 20% time that kind of got copied and then it didn't. But like, how do you kind of leave that time for exploration and delivering for the stuff you know customers want and may have committed?
[0:17:55] We lean pretty heavily on team rotations. And so often I found that it's actually easier to have like a group of people that's just in exploratory mode because it's easier to get into that flow, right? If you're sort of trying to split your own time, you know, 80%.
[0:18:11] Versus 20%, like sometimes that works. And I think that is true for some employees. But I think in the majority of cases, we're like, this is purposely an exploratory team, right? You're doing exploratory efforts. This is a team that's building a product or a research workflow that is pretty well defined and they're executing on it. It's not necessarily any less exciting because there's a lot of work that needs to be done. But oftentimes people enjoy going between those two different states of work and flow. And I think it's also a more interesting way for employees to learn.
[0:18:40] Because they both get to see the value of exploration and the impact that that exploration can later have on our customers. We've talked a lot about AI. Looking ahead, what's something about the future of AI that you think is widely misunderstood today? I think the degree to which the rollout and implementation of different use cases and even across different sectors is going to be very uneven.
[0:19:07] So I think AI, in some pockets of the world, it's impossible to imagine certain companies not leveraging AI today. But I think there's a whole swath of the world that has never used AI, has never thought about how to use AI, and might not for many more years. I think something that's just really interesting to me is thinking about how will this kind of permeate.
[0:19:30] Like, I think...
[0:19:42] I don't think anyone can really predict sort of the ways that the technology is going to be adopted or the places or how or when. And I think that's going to be a really like interesting and like maybe a little bit chaotic period, just as sort of different parts of the world start to leverage AI more or less. Any surprising places that today leverage it more than maybe the you of five years ago would have guessed? I've been surprised the degree to which...
[0:20:07] Many enterprise customers have gotten on board with AI and implemented it very quickly. I think there's sometimes a little bit of
[0:20:15] What do you kind of attribute that to?
[0:20:44] The power of the technology is quite hard to miss. And I think for many of the most successful businesses in the world, they've gotten there by seeing what the next wave is going to be. And I think it's quite obvious that AI is a big deal. And I think the ability for these organizations to say, wow, this is actually really important and we need to find a way to integrate this. Even if we don't exactly know how, a company is going to take us through the process of doing this
[0:21:13] Driving that kind of top-down change from a business perspective isn't easy. But I think when there's enough institutional will, I've been so impressed to watch so many of these customers really quickly transform so many of their different business processes and workflows. So we first met probably 10 years ago, like 2014, 2015, sitting here in 2025. If you think about 2035, which kind of feels probably forever away, the way 2014 does now. But what do you think is true of AI in the world then, in 2035?
[0:21:43] I think in 2035, I think the world is going to look really different. I mean, in a lot of ways, of course. But my expectation is just there will be a lot of places where it will be so hard to imagine how this pocket worked without artificial intelligence. Just like today, it's really hard to imagine so many of the baseline technologies that we take for granted from 10 or 20 years ago. If you imagine like the internet and search was not very common 20 years ago, even maybe 25.
[0:22:12] It's sort of impossible to imagine, like, how did people get places, like, without Google Maps? Like, how did you know where to go? I think there's going to be a lot of moments like that in artificial intelligence, where you'll say, like, well, how did I possibly shop for something on the web by myself? Like, obviously an AI assistant is just going to go find, you know, things for me that I want, whatever. I don't know exactly what it'll look like. But I also think there will be a lot of things that look surprisingly the same, in the same way that if you imagine, you know, 20 or 25 years ago, yes, it's much easier for us.
[0:22:41] To find information and to get places. But there's fundamental things about how humans engage and interact together that I don't think have changed, you know, very much at all. And so my expectation is we'll see another round of that. It just might be on a slightly faster cycle than, say, maybe the first round of the internet wave. Makes sense. Somewhat famously, you co-founded Anthropic with your siblings. Yes. Any surprising parts of that?
[0:23:03] Well, I had great inspiration from Stripe because I saw John and Patrick co-found a company together up close. I think I'm really fortunate that Dario and I have extremely complementary skill sets. And so it's actually just been really fun. We've always wanted to work on something together. We work together at OpenAI. But since we were little kids, we've always wanted to be able to build something together. And so for me, it's actually a huge privilege to be able to get to work with him every day.
[0:23:30] I think just our ability to tackle problems together from the left brain, right brain perspective is very unique. And I have seen many great partnerships like that in organizations or businesses before. But there's something very special about someone that you've known your entire life being able to do it with them. Very neat.
[0:23:47] Any good, like, pre-anthropic projects, like whether it's a pillow fort or, like, kind of ganging up on your parents or, like... I wish I had something so tangible. Well, he taught me to drive, actually. That's a good one, yeah. So I feel like that was more one-directional. But what are, like, collaborative projects we did together? We really like to play video games together. So we have had a lot of building that way. That's very neat. You've been watching AI for the last decade. Do you remember a moment where you saw something, maybe on a screen, maybe in real life, and you're like, oh.
[0:24:16] This is the future, like this thing. It's hard to think of a particular moment, but I think probably unsurprisingly, like GPT-2, when I was pretty new at OpenAI, when we were working on that. And I think even though, you know, now looking back on it, it's like hilariously bad. Just seeing an artificial tool.
[0:24:38] Like respond so convincingly on such a wide range of topics, you know, math and writing and, you know, like there was some French in there and we like didn't, it was just like must have been in its training data, but it could like speak some French. It was like pretty crazy at the time to see it, especially like so up close, right? And to say like, wow, like how is this possible that this model is able to do all of these things? I think that for me felt like a very like.
[0:25:07] Wow moment. There's of course been like many more impressive things since then, but I think there was something about that early phase that felt very interesting. What are you reading these days? When I have time. I'm a big science fiction reader, but I've actually been reading a lot about the mid...
[0:25:25] 20th century in Europe, which I'm very curious about, or the sort of World War I, World War II period. I'm reading a book now called The World of Yesterday, which is just very interesting to just kind of think about how quickly things can change, both from a sociopolitical perspective, but also from an economic perspective, a technological perspective. It's wild to me to think that's almost 100 years ago now, but I think...
[0:25:48] Having access to a lot of really interesting source material from that period, I think has been like really interesting to read. I think we both share an interest in East African politics, which is not something I could say to most people I interview. It's very unusual, yes. But if you were, I don't know, for some reason, Anthropic disappeared tomorrow in a good way, and you had a new job advising some East African government on AI, what would you be saying?
[0:26:12] We're actually very interested in working actually in this exact kind of area. Some of our work in beneficial deployments is around, you know, how do we ensure that parts of the developing world are not left behind kind of in this AI transition? And I think what I would say is, number one, there's so many potential applications, just like we say to enterprise businesses, right, of this technology in a way that could be helpful. We should pick one place to start.
[0:26:36] And I think AI can often feel extremely overwhelming, does to us too, right? Of sort of where do you begin? But I think the potential for artificial intelligence, particularly to just revolutionize sort of mass...
[0:26:48] Education is really vast. And I think there could be an incredible opportunity in partnering with governments around the world to figure out ways to just provide access to basic information and more complex information and AI tutors to a really wide range of populations across all different types of languages too, right? These models...
[0:27:09] Speak every language at an incredibly high level of fluency, including dead languages, rare languages. And I think the ability to provide that level of education to people is really unique. And I'm not totally sure that we've yet even started to kind of crack the process of realizing that. Very neat. When you think back since founding Anthropic, what's something that at the time felt like a really big deal? And in retrospect, just kind of ho-hum.
[0:27:38] Well, there's some funny things, like which office we should start in. We grew out of it much faster than we thought we would. So we were trying to think about where to put our home base, and we picked a really nice office, but we grew out of it in about a year and a half. I think...
[0:27:53] Probably, I don't want to say it's ho-hum, but I think we thought a lot about the incorporation question. So like, should we be an LLC? Should we be a C corporation? Should we be a nonprofit? Should we be a PBC? In retrospect, it was so obvious that PBC was the right decision for us, but it took us like a month or two of discussion before we like formally decided what to incorporate as. And now looking back, it doesn't feel like that should have been a difficult.
[0:28:18] What about the flip side? So something that seemed kind of ho-hum at the time and then in retrospect was actually kind of an important door to walk through. I mean, interestingly, I think our...
[0:28:34] How we structured like our hardware, like it's more of a technical thing, but like how did we decide to sort of, you know, start on one platform versus another and sort of, you know, scale up, you know, our work there. I actually think we made great decisions, but I think we were sort of like, okay, we'll just like start, you know, building on this thing. And I think that just the sort of rate at which that technology as part of the stack has developed, you know, so quickly, I don't know that we like thought about it that hard in like kind of our initial.
[0:29:04] Well, it has been wonderful talking to you today. Thank you so much for the time. Thank you. It was great to be here. Thanks so much for watching. If you're interested in more, you can find us at Vanta.com and Vanta across social media.