The Search Engine for the Agentic Web — Will Bryk, Exa
Exa’s search API is shaped around agent constraints: low latency, narrow token extraction, structured results, and per-customer controls rather than a human search-results page.
Exa says its agent-focused search serves **more than 5,000 companies** and **400,000 developers**. The API can return documents, structured fields, or a compact extraction such as the most relevant 100 tokens from ten results.
Treat retrieval output as part of the agent interface: tune latency, domains, time windows, excluded page types, and returned token volume for each workflow. Voice agents and research agents should not inherit the same search configuration.
Exa says its agent-focused search serves **more than 5,000 companies** and **400,000 developers**. The API can return documents, structured fields, or a compact extraction such as the most relevant 100 tokens from ten results. Treat retrieval output as part of the agent interface: tune latency, domains, time windows, excluded page types, and returned token volume for each workflow. Voice agents and research agents should not inherit the same search configuration. The scale figures, quality claims, and forecast that AI searches will surpass human searches in 2026 are company assertions without supporting methodology in the talk. More search volume also does not guarantee accurate or trustworthy evidence.
This makes retrieval configuration part of an agent’s task-specific interface: result shape, token budget, latency, domains, dates, and exclusions should vary by workflow. It strengthens the case for evaluated, swappable search backends, while leaving Exa’s scale, quality, and growth claims unverified.