Microsoft rolls out Copilot Cowork and advanced research tools: Here’s how they work

Microsoft introduced several upgrades to Microsoft 365 Copilot on March 30, with a new feature for research, Copilot Cowork roll out, and another tool that lets users compare different AI models side by side.

In a blog post, Microsoft said that “all of this innovation is part of Wave 3 of Microsoft 365 Copilot,” which aims to improve its AI offering. Here’s everything you need to know about the latest features coming to Microsoft 365 Copilot.

Copilot Cowork

Microsoft debuted Copilot Cowork early on March 9, and now has finally started to roll out to early-access customers through the Frontier program. This agentic AI feature for Microsoft 365 Copilot is built on technology similar to Anthropic's Claude Cowork.

It can autonomously plan and execute tasks across apps like Excel, Outlook, Teams, and PowerPoint. With Copilot Cowork, users simply need to describe the task, and it will take action while also showing the progress.

New features in Researcher

Microsoft announces new features to Researchers, which are built on multi-model intelligence. While it's already capable of managing complex questions, the new Critique feature will take one step further.

This feature uses multiple AI models from companies like Anthropic and OpenAI, and splits tasks into two steps. “It separates generation from evaluation and utilises a combination of models from Frontier labs, including Anthropic and OpenAI,” the blog post stated.

One model takes on the generation phase and creates the first draft, whereas the second model reviews it and provides refinements. Therefore, it separates writing from checking for improved results. In the DRACO benchmark, which measures accuracy, completeness, and objectivity in research, the Researcher scores 13.8% higher.

Critique is rolling out as a default feature in Researcher and is available when Auto is selected in the model picker.

Model Council in Research

Microsoft has also released Model Council in Research that provides a side-by-side comparison of multiple AI models. The feature can run models from Anthropic and OpenAI at the same time, with each model creating a full report on the same topic.

Once the reports are generated, a third “judge” model steps in to provide a detailed summary on key findings, such as where the models agree or disagree, where they offer unique insights, and others.