Building the Anthropic Team: Daniela Amodei on Hiring 10x AI Engineers | Frameworks for Growth

**Daniela Amodei** (0:00)
At the end of his talk, he said something like, and things are never gonna be chill again. And now we have these cute sweatshirts, and it says things will never be chill again.

**Christina Cacioppo** (0:17)
Welcome to Frameworks for Growth. I'm Christina Cacioppo, co-founder and CEO of Vanta. And I'm here with Daniela Amodei, 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, 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.

**Daniela Amodei** (0:43)
Thank you so much for having me.

**Christina Cacioppo** (0:45)
So Anthropic describes itself as an AI safety and research company that's getting different purposely than a standard technology company. What is that going to mean to you all in practice?

**Daniela Amodei** (0:56)
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 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. In terms of 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. 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. 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.

**Christina Cacioppo** (2:21)
It's very neat. And you're going to touching on some of this, but Anthropic has 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 does that, how do you balance that tension?

**Daniela Amodei** (2:43)
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 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. 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 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 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.

**Christina Cacioppo** (4:15)
Makes a lot of sense, and we use a lot of that at Vanta too.

**Daniela Amodei** (4:18)
That's great.

**Christina Cacioppo** (4:19)
Glad to hear.

**Daniela Amodei** (4:19)
Happy customer.

**Christina Cacioppo** (4:20)
Yeah, very much so. Actually, curious to talk about public benefit corporations. So, most startup founders do not decide to be public benefit corporations, especially on incorporation.