Microsoft Brings Copilot Cowork to Frontier With Multi-Agent AI Workflows

Microsoft has introduced a major upgrade to its AI productivity suite with the launch of Copilot Cowork, a new agentic feature inside Microsoft 365 Copilot that signals a shift toward collaborative “AI teams” rather than a single assistant.

The feature is currently available through the Microsoft 365 Frontier program, which gives early access to experimental tools. Copilot Cowork blends Microsoft’s own AI systems with technology inspired by Anthropic’s Claude Cowork, marking a multi-model approach to enterprise AI.

Copilot Cowork introduces background AI task execution

Copilot Cowork allows users to assign tasks that run independently in the background, without requiring constant interaction.

Users can launch multiple AI-driven workflows at the same time and monitor them through a dedicated dashboard. This setup moves away from prompt-by-prompt interaction and toward persistent AI agents that handle ongoing work.

The system focuses on productivity scenarios such as research, content generation, and data synthesis, all handled asynchronously.

Research agent and structured output improvements

Microsoft also upgraded its researcher agent, which now leverages models from both OpenAI and Anthropic.

The agent generates structured, well-cited reports and can combine insights from multiple sources into a single output. This aims to improve reliability and reduce hallucinations in long-form responses.

To further enhance quality, Microsoft introduced a new Critique feature that separates content generation from evaluation. Two different models work together, one produces the answer while the other reviews and refines it.

Model Council compares AI outputs side-by-side

A new Model Council feature takes this concept further by running multiple AI models in parallel on the same query.

The system compares outputs side-by-side and uses a dedicated “judge” model to summarize key similarities and differences. This approach gives users more transparency into how different models interpret the same task.

Microsoft says this method helps users make more informed decisions and improves trust in AI-generated content.

Performance gains and competitive positioning

According to Microsoft, these upgrades deliver measurable improvements.

Copilot Cowork achieved a 13.8% higher score on the DRACO benchmark, which evaluates accuracy, completeness, and objectivity. The company claims the system now outperforms competing tools such as Gemini, Claude, ChatGPT, and Perplexity in this evaluation.

Availability and future rollout

The Critique and Model Council features are already live within the Frontier program. A broader rollout is expected at a later date, though Microsoft has not shared a specific timeline.

With Copilot Cowork, Microsoft is clearly shifting toward a model where multiple AI systems collaborate rather than relying on a single assistant.

This direction aligns with broader industry trends focused on orchestration, verification, and parallel reasoning across models.

In related updates, Microsoft has also upgraded Microsoft 365 Copilot declarative agents to GPT-5.2, while continuing efforts to reduce Copilot-related clutter in Windows 11.

Via Neowin