Microsoft Launches Copilot Cowork Wave 3 with Multi-Model AI Agents and Critique Function

Microsoft has launched Wave 3 of Microsoft 365 Copilot on March 30, 2026, marking a significant evolution in enterprise agentic AI. The centerpiece is Copilot Cowork, a feature that enables AI to handle multi-step tasks autonomously across Microsoft 365 apps including Outlook, Teams, and Excel.

Cowork builds on Anthropic Claude integration and allows the system to access and output files, manage calendars, and provide daily briefings. It is available through the Frontier early access program.

The Researcher agent received a major upgrade with a new Critique function. One AI model (from OpenAI) generates a draft, while a second model (from Anthropic) reviews it for accuracy and citation quality. Microsoft reports this dual-model approach lifted the Researcher score on the DRACO benchmark by 13.8 percent, claiming best-in-class deep research performance that outperforms Perplexity with Claude Opus 4.6 by 7 points.

Additionally, a new Model Council feature lets users compare answers from different AI models side by side, seeing where they agree or differ. This represents a significant shift toward multi-model orchestration in enterprise settings.

The update signals that enterprise AI is moving beyond single-model chatbots toward sophisticated multi-agent architectures where different AI models collaborate, critique each other, and handle complex workflows autonomously. Microsoft is positioning Copilot as the primary interface for agentic AI at work.