Source: SPEEDA Edge research

Background

Questions

1. Tell us about Exa—what was the key insight that led you to start the company and set out to redesign search for AI agents?
2. What are some of those more complex queries you're thinking about?
3. Can you discuss the structural aspects that prevent Google from effectively handling those types of queries? Then, compare that to what's happening with Exa that makes those kinds of queries more manageable.
4. I don't know if this was recent, but I recently noticed "keyword" as an option in Exa. Are you using this kind of more keyword-based search to address queries for which maybe embeddings don't work as well as keywords? Or is it something else entirely?
5. I'm curious about who you initially had product-market fit with. Who was really using Exa and incorporating it into their products? Can you tell me about that?
6. The first iteration of ChatGPT with web browsing access was somewhat disappointing in terms of the quality of information it retrieved. It seems to have improved a bit since, but overall, it still leaves something to be desired. What’s the challenge for companies like OpenAI and Anthropic here? Do you see them as potential competitors or customers or both in the future?
7. To go back to something you said about how Exa works: Is there a manual process for deciding which pages to turn into embeddings? How does that work? I'm assuming you're not turning every page that's out there into embeddings.
8. What is the criteria for quality? I had the impression it was something to do with the number of shares or something along those lines. Is that roughly correct?
9. Exa has a web interface, but overall, the company comes across as API-first with the web interface as more of a demo of the API. Perplexity seems almost like the reverse—how do you think about that challenge of building both a prosumer product and API at the same time?
10. You recently updated your pricing, making Exa cheaper to use. How do you think about margins and building a sustainable business model at scale?
11. Web search was monopolized by Google in the early 2000s. Gradually, it became fragmented across platforms like YouTube (videos), TikTok (short form), Yelp (restaurants). Now, we’re seeing it fragment across use cases with Perplexity (complex queries) and Exa (finding the best content). Will there be a Google of “search for AI agents” or do you see it fragmenting similarly, and how?
12. Google’s search dominance made it the primary way that people discover new content online—and spawned the birth of an entire industry devoted to getting your content seen better in SERPs. How do you think about the shift in attribution and discoverability with AI agentic search vs. traditional search? How do content creators and publishers have to think differently about AI search?
13. What are the technical challenges that need to be solved for you guys to accomplish what you've been talking about? For example, scaling to the rest of the web and supporting millions of AI agents that are potentially using search?
14. Cool. If everything goes according to plan for you guys over the next 5 years, how do you see Exa looking? And how will the world be different?