Will Bryk

Will Bryk is the co-founder and chief executive of Exa, which builds search infrastructure that helps AI agents retrieve precise, useful information from the web. His central insight is that systems capable of processing thousands of documents and conducting autonomous research require a search engine designed for machines, not human browsing habits.

Bryk grew up in New York City and studied computer science and physics at Harvard, where he researched human-AI interaction, led the robotics club, and built an early search engine. He became one of the first engineers at Cresta, working on real-time AI products, while independently researching the history of civilization sharpened his interest in finding reliable information online.

In 2021, he founded the company with Jeff Wang, his former Harvard roommate, and joined Y Combinator’s Summer 2021 batch. GPT-3’s ability to interpret nuanced requests persuaded him that neural networks could make web search substantially more precise. The startup, originally called Metaphor, developed transformer-based retrieval models, embeddings, training datasets, and GPU infrastructure before releasing its first search engine in late 2022. Bryk renamed the company Exa in January 2024.

Building search for machines

• Search built for AI agents: Bryk designs retrieval around detailed queries, extensive contextual information, configurable filters, and large result sets. Agents need dependable access to current web information because language-model parameters cannot contain the entire, continuously changing internet.

• Neural search and link prediction: Exa represents webpages with embeddings that capture meaning beyond keyword matches. Its early models learned from the text surrounding hyperlinks to predict relevant destinations, making nuanced searches possible even when a page does not contain the exact search terms.

• Retrieval that improves with compute: Applying the Bitter Lesson to search, Bryk emphasizes systems that improve with additional training data and computation. His description of Exa’s architecture outlines distributed crawling, document processing, specialized embedding models, and a custom vector database.

• Complementary retrieval methods: Bryk distinguishes low-latency Exa Search from Exa Websets, which supports more exhaustive investigations. His AI Engineer demonstration combined semantic search to identify relevant engineers with keyword search to locate their GitHub profiles, illustrating how agents can select the appropriate retrieval method for each step.

In August 2026, Bryk reported that Exa served 80 billion pages and tracked 1.4 trillion URLs, emphasizing the need to filter low-quality material before it reaches AI systems. That month, he also announced Jeff Pinner as chief technology officer, reflecting Exa’s expanding infrastructure and research ambitions.