Microsoft's New Work IQ Engine Speeds Copilot Agents 4x, Cuts Token Use 30%

Microsoft has announced a technical upgrade for AI agents such as Copilot in knowledge-intensive enterprise scenarios. Without upgrading the underlying large language model, the response speed of AI agents has increased fourfold, while token consumption has decreased by about 30 percent.

According to Microsoft, project information in enterprise office environments is often scattered across emails, Teams meeting recordings, and technical documents. Previously, AI agents had to retrieve and process these fragmented sources one by one, resulting in high latency and computing costs.

The new solution is built on SharePoint and uses Microsoft's new Work IQ knowledge compilation engine. The engine automatically converts scattered information across an organization into a structured knowledge base that AI agents can directly and efficiently access, reducing repeated retrieval and processing.

Use cases include cross-document Q&A, team context retrieval, meeting note summarization, project status aggregation, internal process lookup, customer communication material preparation, and drafting initial content based on existing materials.

The feature is now available to Frontier Preview tenants, but only to customers who have enabled pay-as-you-go billing. Organizations can add the knowledge base to existing SharePoint sites, including sites supporting Teams channels and Microsoft 365 groups. Administrators can manage access permissions and spending through the Microsoft 365 admin center. Microsoft also plans to extend the capability to custom agents in Copilot Studio and develop automated maintenance so that the knowledge base stays updated with enterprise content.