Language model agents improve search by customizing their interfaces
Let Agents Design How They Search: Automatic Interface Optimization for Agentic Search
Computation and Language
Summary
Searching information with AI assistants depends a lot on how they interact with the data they explore. The authors developed a system called SearchWeaver that lets the AI change and improve its own search tools and instructions, making searches more accurate and useful. This approach means the AI figures out the best way to use the data and tools for each specific collection of information. Tests showed that this method improved the quality of search answers and ranking compared to usual ways. This shows that letting AI tailor its search setup improves its performance.
What this means in practice
- •For software engineers: Improve AI-powered search tools by enabling automatic optimization of search interfaces tailored to specific data collections.
- •For digital content managers: Enhance content retrieval quality by adapting search interactions automatically to different document collections.
Authors
Sangjune An, Jaeyoon Jung, Yumin Choi, Jinheon Baek, Junmo Cho, Sung Ju Hwang, Hae Beom Lee
Abstract
Agentic search depends on the interface through which a language-model agent accesses and explores a corpus. Existing interface designs are fixed by hand and applied unchanged to every corpus, although effective designs may depend on the corpus. We introduce SearchWeaver, a framework for automatic search interface optimization guided by downstream search performance. A proposer uses execution trajectories and task rewards to jointly revise corpus artifacts, executable tools, and usage instructions. This pro cess optimizes what information is prepared in advance, which operations tools perform, and which decisions remain with the search agent. With GPT-5.4-nano, the optimized interfaces improve answer accuracy from 71.6% to 78.4% on BrowseComp-Plus and average nDCG@10 from 45.7 to 55.0 across four BRIGHT subsets over Direct Corpus Interaction. These findings highlight the value of jointly optimizing corpus artifacts, executable tools, and usage instructions to develop effective search interfaces.