Microsoft AI Tools Achieve 400% Speed Increase, Reduce Token Consumption

Microsoft has announced a significant technological advancement for its AI agents, such as Copilot, intended for enterprise knowledge-intensive scenarios. This new technology, developed without upgrading the foundational large models, reportedly increases AI agent response speed by up to four times and reduces token consumption by approximately 30%.

Previously, accessing and processing fragmented information scattered across emails, meeting records, and technical documents in corporate environments led to high latency and computational costs for AI agents. This often required agents to individually retrieve and handle diverse data sources.

The breakthrough utilizes Microsoft's "Work IQ knowledge compilation engine." This engine automatically transforms dispersed information from across an organization into a structured knowledge base. AI agents can then directly and efficiently query this compiled knowledge base, improving performance for tasks like cross-document question answering, team context retrieval, and summarization.

This feature is currently rolling out to Frontier Preview tenants. Organizations can add these knowledge bases to existing SharePoint sites, which also support Teams channels and Microsoft 365 groups. Microsoft plans to extend this capability to custom agents via Copilot Studio and develop automated maintenance features to keep the knowledge bases synchronized with evolving company content.