Will Bryk explains Exa's effort to turn web search into information infrastructure for AI agents. He contrasts recommendation-style results with complex queries that require a precise set of matching people, companies or documents, and describes embeddings as one part of a broader retrieval system rather than a complete solution.
Will Bryk outlines tradeoffs between fast searches and slower exhaustive research, along with token extraction and structured output for downstream agents. He presents domain and time-window controls as ways to adapt retrieval to different customers. Performance and adoption figures are company-reported claims in the talk, not independently verified comparisons.
Will Bryk describes combining public-web retrieval with licensed data-provider access and a marketplace in which developers choose useful information. His longer-term goal is near-complete, trustworthy information for agents and people. Forecasts about agent search volume and social coordination are presented as ambitions and expectations, not established outcomes.
Watch on YouTube
