Sponsored by: You.com | Achieving State of the Art on Deep Search: Lessons for Agent Builders

AI agents need to go beyond simple lookups. They need to search deeply, synthesize across hundreds of sources, and verify facts across multi-step reasoning chains. But building a deep search system that's accurate, fast, and reliable at scale is a fundamentally different problem than traditional software systems. Saahil Jain, CTO of You.com, shares how the team built the Research API — ranked #1 on DeepSearchQA with 83.67% accuracy and 93.16% F1 — and what they learned along the way. The talk covers: why a robust but simple harness paired with high-quality search tools outperforms complex agent orchestration; the architectural choices behind pushing the accuracy-latency frontier across benchmarks from SimpleQA to DeepSearchQA; and why single-run eval numbers are misleading for agents, with practical techniques for measuring what actually matters. Built for builders on platforms like Databricks who want to add deep search to their agentic workflows.