Microsoft has made its new Copilot Cowork capability available to participants in its Frontier program, extending the role of Microsoft 365 Copilot beyond content generation into task orchestration and execution across enterprise applications.
The update, outlined in a blog post by Jared Spataro, Chief Marketing Officer, AI at Work, Microsoft, forms part of a wider set of Wave 3 enhancements focused on multi-model AI systems and long-running workflows. Microsoft has integrated the technology that powers Anthropic’s Claude Cowork, enabling long-running, multi-step work across enterprise applications.
Copilot Cowork is designed to allow users to specify an outcome, after which the system generates a plan, coordinates tasks across tools and files, and advances work with human oversight. Spataro described this as a move toward AI that can carry out connected sequences of actions rather than respond to individual prompts.
“Describe the outcome you want, and Copilot Cowork creates a plan, reasons across your tools and files, and carries work forward with visible progress and opportunities to steer.”
The capability draws on Microsoft’s broader “multi-model” approach, combining internal AI systems with models from external partners such as OpenAI and Anthropic. This allows different models to contribute to various stages of a workflow, according to Microsoft.
Multi-model AI Enhances Research Accuracy and Evaluation Workflows
Alongside Cowork, Microsoft has introduced updates to its Researcher tool using multi-model intelligence. Researcher helps users synthesize information across sources, generating comprehensive analysis with cited, reasoned responses. The “Critique” function separates content generation from evaluation, using a combination of models from Frontier labs including Anthropic and OpenAI. One model produces an initial response, while another reviews and refines it before delivery.
This approach improves Researcher’s results on Microsoft’s DRACO benchmark, which measures research quality in deep research accuracy, completeness, and objectivity, by 13.8 percent, Spataro noted.
In addition, a “model council” function allows users to compare outputs from several different AI models side by side providing transparency into AI reasoning and helping users make more informed decisions. Microsoft CEO Satya Nadella wrote on LinkedIn:
“You can run multiple models on the same prompt at the same time, so you can see where they align and diverge, and understand what each adds.”
That improved transparency could help reduce blind spots and increase confidence in AI-driven workflows.