BRK309 Breakout Wed 18 Nov 13:45 PST
Optimizing AI agents to meet your goals: Grounding with Web IQ
Knut Risvik, Husein Sharaf, Ray Zhong
Summary
From the recording
Microsoft said Web IQ Auto, announced the previous day, is generally available on Azure. It moves search orchestration, including fan-out and iteration, out of the developer's agent loop and into Web IQ. In Microsoft's demo, searches fell from 35 to 7 and tokens from 370,000 to 25,000, a 93% reduction. Microsoft will open source its freshness benchmark harness and data. Caveat: figures come from Microsoft's own demo on selected queries; pricing for Auto and regional availability were not stated.
Session description
Microsoft's catalogue description
Every grounding call is a trade-off. The highest-quality answer isn’t always the right one — sometimes your agent needs speed, sometimes freshness, sometimes precision at any cost. This session breaks down the quality-latency-cost frontier behind agent retrieval with Web IQ: when to let auto routing decide, when to orchestrate yourself, and how to design grounding strategies that hold up in production. Leave with a decision framework for any agent you’re building.