The Search Engine for the Agentic Web — Will Bryk, Exa
Exa's CEO, William Bryk, discusses the impending AI search boom, predicting machines will soon outpace human searches dramatically. He explains Exa's evolution from a consumer search engine to an API for AI agents, driven by the realization that LLMs need external, 'perfect' retrieval to overcome the limitations of traditional search and combat societal misinformation. Exa leverages neural networks, embeddings, and customized features to provide fast, accurate, and cost-efficient information, with plans for a private data marketplace, aiming to deliver 'a year of research in a second' for AI agents and foster a deeply informed society.
William Bryk, CEO of Exa, discusses the impending paradigm shift in web search, predicting that AI systems will conduct more searches than humans by 2026, eventually outnumbering human searches by a factor of 1,000. Exa positions itself as the “search engine for AI agents,” serving over 5,000 companies and 400,000 developers, including coding agents like Cursor and go-to-market agents like HubSpot.
The Problem: Misinformed Anarchy and Flawed Search
Bryk highlights the internet’s vastness—trillions of pages—making it impossible for humans to achieve deep understanding. Traditional search engines like Google, while effective for simple queries (e.g., “Costco homepage”), function more as recommendation engines than precise information databases. An example is searching for “shirts without stripes” and receiving results with stripes. This limitation means Google cannot handle complex, granular queries such as “find me everyone in Singapore who works on AI search and any blog post or research paper they’ve written.” This leads to a world where individuals lack complete and trustworthy information, a state Bryk terms “misinformed anarchy.” He views solving this information deficit as “the most important neglected problem,” warning that failure to do so could lead to a dystopian future where societies are easily manipulated.
The Thought Experiment Behind Perfect Search
The inspiration for Exa emerged in 2021 with the advent of powerful transformers (like GPT-3). Bryk’s foundational thought experiment involved taking a complex query and a document, running an advanced language model over them, and asking if they matched. This process, while highly accurate, would cost approximately $10 million per query if applied to a trillion documents. The core challenge for Exa became reducing this cost by nine to twelve orders of magnitude.
Exa’s solution relies on neural networks and embeddings. While running a full neural network over every document for every query is infeasible, pre-processing documents into intelligent structures like embeddings allows for efficient and accurate matching. This approach enables Exa to handle queries like “shirts without stripes” effectively, essentially “stacking more layers” to improve comprehension.
Exa’s Evolution and Business Model
Initially named Metaphor, Exa launched its consumer search engine in 2022. The release of ChatGPT two weeks later fundamentally altered their trajectory. The company received unexpected Twitter requests for API access, including from Bryk’s own roommate. This led to the realization that while Exa was building the “wrong product,” it was for the “right reason.” The insight was that even colossal models like GPT-5 are tiny compared to the internet and will always require external retrieval. AI agents, unlike humans, perform complex, truth-seeking searches and prioritize accuracy over SEO or ads.
This shift led Exa to become an API-first company, developing a specialized search engine for AI agents that aligns with their original mission of “perfect search.”
Current Capabilities of Exa
Exa offers several key features tailored for AI agents:
- Complex Queries: Ability to handle highly specific, multi-faceted queries (e.g., “every startup funded by YC working in AI, their batch, and status”).
- Fastest Search API: A 200-millisecond search endpoint crucial for real-time AI interactions like voice agents, where every millisecond counts.
- Efficient Token Extraction: Provides only the most relevant tokens (e.g., 100 tokens from 10 documents), significantly reducing downstream LLM costs.
- Structured Output: Delivers specific data points (e.g., college, graduation year, most cited paper for an engineer) in a structured format, simplifying data consumption for agents.
- Customization: Exa’s system is highly flexible, allowing each customer to effectively have their own “search engine” tailored to specific needs, such as domain restrictions, time windows, or exclusion of product pages. This ensures optimal quality for diverse AI agent tasks.
Exon: A Marketplace for Private Data
Recognizing that AI agents need access to both public and private data for comprehensive truth-seeking, Exa launched Exon. This platform enables data providers to monetize their proprietary data by making it accessible to developers building AI agents. It creates a “free market” where data providers set prices and developers choose valuable datasets, facilitating queries that combine public web information with private data (e.g., website visitors from SimilarWeb).
The Vision: A Year of Research in a Second
Bryk’s ultimate vision for Exa is to achieve “perfect information” for AI agents, akin to performing a “year of research in a second.” This means providing instant, comprehensive answers to any query, regardless of complexity. He believes this infrastructure is critical for societal well-being, particularly in an age where misinformation can have profound impacts, citing the 2028 presidential election as an example where informed citizens are paramount. The exponential increase in AI-driven searches—each grounding itself in search—will necessitate this level of accurate, perfect information to prevent societal coordination problems and foster deep understanding globally.