Microsoft Launches Copilot Cowork and Critique: GPT and Claude Now Collaborate in Microsoft 365

Microsoft has taken a major step in its AI strategy by launching two new features for Microsoft 365 Copilot: Copilot Cowork, an agentic AI that handles complex multi-step workflows autonomously, and Critique, a multi-model system that pairs GPT with Claude to cross-check responses, boosting accuracy by 13.8%.

Key Highlights

• Copilot Cowork manages multi-step tasks across Excel, Teams, and SharePoint without constant human oversight
• Critique pairs GPT (generation) with Claude (verification) in a cross-review system
• Model Council enables multiple AI models to contribute perspectives to a single decision
• 13.8% improvement on the DRACO benchmark compared to standalone competing tools
• Available through the Frontier early-access program for enterprise customers

Copilot Cowork: Autonomous AI Agents in Microsoft 365

Copilot Cowork is designed for long-running, complex tasks that require coordination across multiple Microsoft 365 applications. Users describe a task, and Cowork creates a plan, reasons across available tools and files, then executes the work step by step with visible progress tracking.

Practical use cases include monthly budget reviews, calendar management, and cross-team task coordination. The system operates under the Work IQ framework, which grounds AI responses in organizational data while maintaining security and human oversight throughout.

Barton Warner, Senior VP at Capital Group and an early adopter, emphasized that Cowork is about "taking real action" by "connecting steps, coordinating tasks, and following through."

Critique: When GPT and Claude Check Each Other's Work

The most innovative feature is Critique, embedded in Copilot's Researcher agent. The concept is simple but powerful: GPT generates an initial response, then Claude reviews it for accuracy and citation quality before presenting it to the user.

This cross-review mechanism, which Microsoft calls the "Model Council," improved research benchmark scores by 13.8% on the DRACO industry measure, outperforming standalone tools from Google, Perplexity, and individual models.

Nicole Herskowitz, corporate VP of Microsoft 365 and Copilot, stated: "Having various models from different vendors in Copilot is attractive, but we're taking this to the next level where customers get the benefits of the models working together."

Microsoft plans to make this workflow bi-directional in the future, allowing GPT to review Claude's drafts as well.

A Strategic Shift Toward Multi-Model AI

This launch marks a strategic turning point for Microsoft. Rather than relying exclusively on its OpenAI partnership, the company is now explicitly integrating Anthropic's models into its flagship products. This "multi-model by design" approach responds to growing enterprise demand for flexibility across AI providers.

The challenge remains significant: only 3.3% of Microsoft 365's 450 million commercial users currently subscribe to paid AI features. Copilot Cowork and Critique represent Microsoft's bid to deliver compelling enough value to accelerate enterprise adoption.

Availability

Both features are currently accessible through Microsoft's Frontier early-access program for enterprise customers. No general availability date has been announced yet.

Source: Microsoft 365 Blog