Microsoft debuts Agentic Retrieval Preview in Azure AI Search to handle complex queries better

Microsoft has introduced agentic retrieval, a new feature in Azure AI Search designed to improve how AI handles complex, multiturn questions. This update uses a smarter approach that plans and runs multiple focused searches instead of a single broad query.

Microsoft says this can boost answer relevance by up to 40% compared to traditional methods. Agentic retrieval is now available in public preview.

How agentic retrieval changes AI search

Agentic retrieval uses an Azure OpenAI model that reviews the entire conversation history to understand what users are really asking. It breaks down complex questions into smaller, targeted subqueries.

These run in parallel, searching both keyword text and vector-based semantic data at the same time. This hybrid method improves the search’s recall and accuracy.

After gathering results from all subqueries, Azure AI Search reranks them into one clear response. The system also provides a detailed activity log showing how each step was handled. This makes it easier for developers to see how answers were generated and troubleshoot if needed.

Also read: Azure Sphere May update brings tighter security and control for Legacy access

Early feedback and availability

AT&T is among the first to adopt agentic retrieval in their enterprise AI solutions. Mark Austin, VP of Data Science at AT&T, said the technology meets the speed and complexity requirements of their applications

Agentic retrieval is currently in public preview in select regions. Pricing covers token usage for query planning via Azure OpenAI and semantic ranking through Azure AI Search.

Also read: This is how much Walmart spent on Microsoft Azure cloud services

Semantic ranking is free during the preview, with normal charges starting later. Developers interested in smarter AI-powered search can now explore this new capability.