The Search Engine for the Agentic Web — Will Bryk, Exa
Source
youtube.com
Author
AI Engineer
Date
Context
Exa frames search for AI agents as a problem of satisfying detailed requests at web scale. Its search-engine talk uses a simple thought experiment: a LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → can assess whether one document fits a complex query, but repeating that assessment across a trillion documents for every search would be prohibitively expensive.
That framing separates the desired judgment from the cost of delivering it. The talk also argues that mainstream engines are built to recommend rather than answer, so a request for shirts without stripes can return striped shirts, while an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → needs its exclusions and constraints honored. The search service therefore has to preserve those requirements while finding relevant material efficiently.
Key takeaways · AI-distilled
Exa founder Will Bryk says 2026 is the first year machine-issued web searches will exceed human-issued searches, and he expects a thousandfold gap between the two going forward.
Bryk uses a thought experiment: checking a query against a trillion documents with a language model would cost roughly $10 million per search. He frames Exa's engineering problem as reducing that hypothetical cost by many orders of magnitude.
Bryk argues mainstream search engines are built to recommend rather than answer precisely, citing that searching 'shirts without stripes' still returns striped shirts, which fails the literal, exact-match needs of AI agents.
Exa started in 2021 under a different name, launched a consumer search engine in 2022, and only built an API after unsolicited requests, including one from a Twitter user and one from a roommate, revealing unmet demand for programmatic access.
Terms in this piece · Glossary
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
videoVoice AI Where One Wrong Question Breaks the Rules - Shiv Indap, Upstart & Brooke Hopkins, Coval
video78% Wanted This AI Feature. Almost Nobody Used It - Supriya Subudhi, Bumble
videoWhen Running Your Own AI Models Beats Paying for APIs - Steven Latré, imec
videoHow to Decide What Your AI Agent Should Remember - Wojciech Pluta, Oracle
