Building AI Search Engine for the GPT-5 Era | Will Bryk, Exa

In 5 years, we could have AGI systems that completely automate all human labor. AIs do all the repetitive work and humans do all the novel work. Every human is going to become a product manager of a team of AIs. How are you supposed to plan for that? I think it's extremely hard to predict where the world will be in 5 years. 1-year plans make sense right now. 3-year plans are really hard and 5-year plans are impossible. So because the AI market is changing so fast, like every month, like new systems come out that make new things possible. The right way of navigating that is to think from first principles about like, what does the market need that still will be true in a year? Because if you're thinking about what does the market need right now, a month later something new is going to come out and they're not going to need it anymore. So you have to think a little more long term. You have to be a little more strategic than in the past. You know, for example, like when I was in college, AI was not at all as prominent as it is today. Everyone today talks about AI, but there were a few people who were aware of what was going on. I was lucky enough to go to a Westworld watching party at OpenAI, and I was actually on a beanbag with Greg Brockman. Just chilling on this beanbag. And we were arguing about the scaling hypothesis. And the scaling hypothesis is this idea that you keep putting more compute into transformers and they'll just keep getting better. And that's how we get to AGI. That was a crazy idea when I was in college. The vast majority of people didn't believe it. I didn't believe it. Greg was arguing that if we just keep scaling these things, we'll get there, we'll get to AGI. And I was arguing that we need new types of methods. I think we were both right in our own way, but we kept seeing progress from GPT-1 to GPT-2 to GPT-3, eventually GPT-4. And at some point along that trajectory, I was like, holy cow, like these systems, when you scale them, they just get really good. I think the people who were earlier to that conclusion did better. And we kind of apply the same scaling logic to Exa. We are building transformer-like systems for search. And we also know that if you keep packing data and compute into the search engine, it will get better and better. That's a hypothesis we have. It's like our own scaling hypothesis for search. It's very different from traditional search engines like Google or Bing, which basically have stayed the same for you know, many years, whereas like Exa is getting better like this. And so we're thinking a lot about like, where is the future going? So we see a world where there are agents everywhere, GPT-5-level AI agents navigating the web, doing all sorts of tasks. This future is coming. They're going to need search, all of them. Hey, I'm Will. I'm the CEO of Exa. We're building the next generation of search. One good way of understanding Exa is in contrast to traditional search. So traditional search engines use mostly keywords. Okay, so if you're using a traditional search engine and you want to find startups working on futuristic hardware in the Bay Area, traditional search engines will use keyword matchings. The results they give you will be documents that contain the words startup and hardware and Bay Area. But startups that are working on futuristic hardware in the Bay Area, they don't typically have those terms. Like you might have a rocket company in SF. A traditional search engine won't be able to find that rocket company, but Exa can because We understand the meaning of documents. We understand the meaning that, oh, if it's a rocket company in SF, then it does match startups working on futuristic hardware in the Bay Area. We believe it's possible to have perfect search over the web, meaning whatever information you want, you get exactly that. We help companies integrate this high-quality knowledge into their applications. So we recently raised $17 million from Lightspeed and NVIDIA. Our revenue is doubling every quarter. We're building the next generation of search. I came into college wanting to study physics to like, I want to understand how the universe works. And I thought physics was the right way to do that. Something big that influenced me was watching The Social Network because I was studying at Harvard and The Social Network took place in Harvard. It was actually a very accurate movie and it was very inspiring to see this guy change the world just on his laptop. I realized that you could have like a massive influence just coding on your laptop. In a way that you couldn't as much with physics. It was very clear that, that, that the AI could understand the problems of physics. So I went into computer science and I think that turned out to be right because now like the AI is getting so good that it should be able to just tell us the answer or infuse it into our brains. Before Exa, on the side, I was writing a history book. I got really excited about world history and I decided I'm just going to write a book about it because I'm like, no one has captured in in a book, my, the level of excitement that I had. And so I was doing a lot of research for the book. I quickly realized that Google is actually not good enough for that type of research. Like Google is great for surface level investigations. Once you start like trying to go deeper and trying to understand any topic deeply on the web, Google really fails. For example, if I want to find like all the research papers on poverty in ancient Rome, it's actually really hard to find that on Google. Not every paper will mention the word poverty. And so I was doing the research for this book and it was really hard to find things. And then at the same time, GPT-3 had recently come out and GPT-3 was this magical creature really that I could talk to and it could understand like anything I say at a very deep, complex level. And so the thinking was, what if we could apply the same technology of GPT-3 to search? What if you can make a search engine that actually understands you at a deep level? And it's been the same goal ever since. The first year and a half of Exo, we did research into search models, into how can we take transformer models and apply them to a search engine. No one's really done that before. It took a long time to figure out how to do it well. And that required persistence. Like we were just banging our head against the wall for a year and a half, trying out different models, trying out different datasets. And eventually we got something that worked really well. If we didn't have the persistence, you know, 6 months in, we might have given up, but we didn't. Early November 2022, we launched the first version of Exa to the public. We built a search engine that was perfect for AI applications. Basically, AI systems, they have all this intelligence, but they're lacking in knowledge. And so when they need knowledge, they go make a call to Exa and get exactly that knowledge. Then ChatGPT came out a few weeks later, and that was a big moment for the information ecosystem. But for us, it was really interesting because we started getting requests for API access to our search engine. We started getting requests for API access, first from a friend who was actually like living downstairs. I told him, no, sorry, we don't have API access. And I didn't really think much of it. But then we kept, then we got a request for API access from someone from some company in Germany. And we also told them, no, sorry, we don't have an API. And then we kept getting requests for API access. And we realized that because of ChatGPT, people were starting to build AI applications all over the place for all sorts of businesses and all these AI applications needed search, like the AIs themselves needed to search. And that's when we started to realize, okay, XLOOK could be really useful for these AI applications. Yeah, so our initial customer found us. And so one lesson there is just like be a really good listener to the market. Like what are people repeatedly saying they need? And you might have some idea of what, you know, what you're going to sell. But then if people keep requesting something like API access, maybe you should start listening to them. Like we cared more about learnings from the customer than getting a lot of revenue. And so yeah, we were like opening our ears to what do the customers need. Over time developed a hypothesis about how we should price and what types of customers we should pursue. I think you can guess where companies like OpenAI and Anthropic are going to build based on like what are big markets that they could tackle. Agents, like automating work is a clear, huge market. And so they're clearly going to do that. Okay, so now you know what types of things they want to do. Can they do it? Well, then you think about the fundamentals of LLMs. You think like, okay, LLMs can, as long as you can make training data for some objective, they will get better at that objective. Can you make training data for agentic behavior? Definitely. You just have a bunch of examples of navigating the web in order to buy plane tickets. And if you have a million examples of that, then now the LLM will know how to buy plane tickets. Yeah, so any sort of tasks such that you could create data for it, LLMs will get near perfect at that task. So that's like a fundamental way of thinking about this. So will AI get really good at navigating the web? Yes, that's pretty easy because it's easy to generate lots of navigating the web data. Will AI get really good at robotics? Yes, but probably on a longer-term horizon because gathering lots of robotic training data is harder because it's hardware and hardware is hard to work with. So what do we learn from that? We learn that AI's navigating the web will come pretty soon. AI's navigating robotics will take longer. So just from like simple first principles, you can like guess where things are going. Picture in your mind how crazy good technology will be and AI will be in a year. And then build for that world.