Inside the Jeff Bezos-Backed Startup Trying to Automate Engineering

0:00This episode is supported by Anthropic, the team behind Claude. Every business wants to move faster. Few can afford to be wrong. That's why the brands you trust also trust Claude.

0:14Welcome to Tech News Briefing. It's Thursday, November 12th. I'm Julie Chang for The Wall Street Journal. Imagine AI stepping out from behind our screens and into the physical world. That's the promise of physical AI, systems designed not to just generate text or code, but to engineer actual machines. And that's what Amazon-backed startup Prometheus is working on, using the technology to automate manufacturing and speed up the way the real world gets built. WSJ reporter Kate Clark sat down with Prometheus co-CEO Vic Bajaj at a recent WSJ tech event to discuss what he thinks is the next frontier in AI.

0:55They talked about the startup's eye-popping$41 billion valuation, what it's like to run a company alongside Jeff Bezos, manufacturing competition with China, and more. Here are some highlights from their conversation.

1:11Kate Clark: So for many years, investors were pouring hundreds of billions of dollars into large language models, chatbots. Now there's a lot of excitement around physical AI, which is the category that you are in. Why do you and others see that as the next frontier for AI? Well, language models are, of course, far more than chatbots, right? We have invented a kind of intelligence that is going to be phenomenally useful. But to engineer the world around us, we have to work with physical systems, even to design the world around us. It's highly multimodal. So that is, first of all, why it requires a new set of technologies.

1:46But, you know, roughly 60 to 70 percent of the world's GDP is actually involved in things that are physical. That can grow even more. It is so difficult to engineer and develop things in the physical world. You know, an aerospace engineer working on something like a turbofan engine may go an entire lifetime, an entire career, before he can probably fly on an airplane that's powered by an engine to which he contributed a small component. And it's remarkable that we can do things like that over and over again, right? In the jet engine case, it's why the world is so much smaller. But what if that person, that engineer, could have that experience in one year rather than once every 20 years?

2:28That kind of compounding productivity will drive a lot of economic growth in a segment of the global economy that's already the lion's share of GDP.

2:39Kate Clark: Have we had the chat GPT moment for physical AI, or what might that look like? So physical AI and the way that you engineer things and interact with manufacturing and the physical world is much more heterogeneous. You know, take that iPad that you're holding. There are manufacturing processes in it that literally date almost from the Bronze Age, things like casting. And there are manufacturing processes that we've invented a mere decade ago that involve rearranging and patterning materials on the scale of a few atoms. All of that has to come together. and it's not reasonable to assume that there would be one interface to that or one universal way to approach it as there is for a chatbot but there are very, very promising things where we are already seeing orders of magnitude acceleration of design cycles that together constitute a chat GPT moment, if you will a set of cases where the entire process from design to validation of that design to manufacturing of that design can happen in silico before it happens in the real world.

3:47We see a lot of examples of that emerging already.

3:50Kate Clark: You have a$41 billion valuation, a staggering valuation for a young company. How do you justify that at this stage? We are very humble about that. You know, it's not something that can be justified at the inception of a project like this. The vision is very large. Figures like that are both staggering. They represent, rather than something that we can justify at the moment, an awesome responsibility to our investors and the broad community that's entrusted us with capital to enact a vision that we think will be transformational. So it's the potential that justifies the vision, but we would never say that it is justified in that sense today.

4:35It's the potential.

4:36Kate Clark: What's it like working with Jeff Bezos? It is daunting at times, but also pretty amazing. I've known Jeff for a long time. We've never worked together closely, but he's invested in companies that I founded, and I knew him when I was at Google. That's when I first met him. Jeff is someone who has built Amazon really with his own hands, and that has given him that experience, and the experience of being really at the frontiers of some very difficult engineering problems at Blue Origin has given him both an instinct for the physical economy, but also a hands-on experience that is extraordinarily useful to us.

5:15So I learn something every day by working with him, particularly on the product side. He's a master of product strategy. And what we're trying to do is probably one of the more unprecedented and complicated product strategies in the field.

5:32Kate Clark: And your co-CEOs, how do you divide up the responsibilities? We don't really divide up the responsibilities in that way. We are all involved in everything, and that has served us well. We debate decisions a great deal together with the teams, and it's actually very comforting to have more minds at work on the kinds of decisions that we're making, and so we try to be involved and contribute different perspectives to everything that we're doing together. Can you take us inside one of those debates? Or is there something that you and Jeff have disagreed on that you could tell us about? Well, maybe someday, not today.

6:12But there are things that, like in all business relationships, I wouldn't say that we've disagreed, but our instincts have been in different directions. And we debate and come to a position that is better than either one of us would have come to alone, which is the whole point of having that kind of relationship. What questions do you have about physical AI? If you're a listener on Spotify, let us know in the comments. Coming up, can China pull ahead in the race to automate manufacturing? That's after the break.

6:59for Anthropic. That's their whole mission.

7:04Kate Clark: Let's talk about China. China has shown it can compete fast despite less access to advanced chips. Are we underestimating how quickly China may be able to pull ahead? This is an interesting question. You know, 15 years ago before I went to Google, I was an academic. I was working on everything from fundamental tests of quantum mechanics, quantum mechanics harnessed in devices, all the way to early detection of disease and life sciences, right? And so I got to know my colleagues in China, including many students. You know, at that time, 15 years ago, there were far fewer barriers. And as I talk to politicians today, I have come to believe that we're systematically underestimating and really misunderstanding the basis of Chinese competitiveness.

7:50Really, many people have a model that's very anachronistic. It's of intellectual property theft rather than domestic innovation. But if you look at it, our isolation of China and their emulation of what has worked well in Western science has actually created an economy of innovation that is far more threatening to us if China becomes an adversary than most people realize. And in all the fields where I have worked in my career, there are centers in China that really achieve a parity with the West. And we have to change the way that we're thinking about that if we want to compete.

8:30Kate Clark: And specifically when it comes to physical AI, there are advantages due to their manufacturing base, right? How do you think about that? You asked about competition. Should a competitor emerge, I would predict that it would be most likely in China. There is an enormous degree of experimentation in China, far more than in the U.S. supply chain. on these topics. It doesn't yet have the organizing force of the hand of central government, but it could. And to see an example of what's possible there, look at what has happened in the semiconductor supply chain in China. Look at the innovation, if people read the Chinese literature, even of alternatives to EUV processes on which frontier semiconductor fabs depend, where in the West, We are working with technologies whose lineage is really in the 1980s, 1985 or so.

9:19The embargoes and restrictions have driven tremendous advantages, right? So physical AI is going to be similar at some point because there's just so much experimentation. There are many models, and there is, ironically, not to the elite researchers at frontier labs, but to the average researcher and average practitioner, there's a lot more compute available.

9:45Kate Clark: So we were just talking backstage about the PR problem for the AI industry. Do you think that Silicon Valley has done a bad job explaining how this new technology can benefit ordinary people? Of course we have. I mean, I think Silicon Valley is, for all its wonders and benefits, it's a very insular place. And we're rightly concerned about the safety of new technologies, right? One should be. But we look upon people who are worried about data centers, and we think sometimes that they're Luddites or they're resisting this technology. It's not the case at all, in my view. We have to empathize with what they're hearing and what they're feeling.

10:26So there are those among us who have spent a few years explaining to people that they will be out of jobs, their children will be out of jobs, their grandchildren will be out of jobs. when actually I think what I'm saying is far closer to the reality and far more consistent with the history of technologies. And now we have told them that there are other dangers, maybe existential dangers. So it's no wonder that people are opposed to data centers because they worry that people who are warning them about these dangers have no optimistic framing and further that those same people are profiting by creating those dangers.

11:07And so there's a trust issue. And I think we ought to admit that, embrace that, and try to really explain what the positive benefits of this technology can be. New technologies cause disruption, but they also create new industries and new prosperity. That's the history of Silicon Valley. We should be proud of that and embrace that, but be more honest and complete in our descriptions.

11:33Kate Clark: We're almost out of time, but I did want to ask you, Prometheus is a loaded name. What are you trying to say by calling the company Prometheus? You know, it was an internal code name for the project, which stuck. I do think it's appropriate. There is a feeling that we are, not we, just in the entire AI community, that we are accessing something that's new for the first time that is a very powerful technology. that we have to engage with the benefits of that and the dangers of that. And often we look to mythology to understand those benefits and dangers. Powerful and in some ways as alien as this technology is, there are many predicates and examples in history that we ought not to ignore about how to deploy such technologies well and how to preserve their benefits and minimize the harms.

12:25That is what the name evokes for me. That was Journal Reporter Kate Clark speaking with co-CEO of Prometheus, Beek Bajaj. You can watch the full chat on wsj.com. And that's it for Tech News Briefing. If you're a listener on Spotify, be sure to leave us a comment. Today's show was produced by me, Julie Chang, with supervising producer Katie Ferguson. We'll be back tomorrow with a new episode. Thanks for listening.

12:59See what's next at claude.ai.