Microsoft Copilot Is Using a Surprising AI Trick to Create More Accurate Research Reports |
Microsoft Copilot Is Using a Surprising AI Trick to Create More Accurate Research Reports
If you’re looking for more trustworthy AI-driven research, check out these new Microsoft features.
BY BEN SHERRY, STAFF REPORTER @BENLUCASSHERRY
Illustration: Inc; Photo: Getty Images
Microsoft has announced some major upgrades to its 365 Copilot platform, including to the latest version of its Researcher agent, and the soft launch of its Copilot Cowork feature, built in partnership with AI titan Anthropic.
In a March 30 blog post, Microsoft said that it added new features to Copilot Researcher, the company’s version of the report-generating deep research agents offered by OpenAI and Anthropic. These agents typically work by developing a plan for how to conduct research into a given topic, then creating multiple sub-agents to check hundreds of sources simultaneously. Copilot’s updated Researcher agent takes a slightly different approach.
Microsoft said that a new “Critique” mode directs the Researcher agent to use two separate models: one to generate the research report, and another to act as an expert reviewer. “By giving evaluation as much emphasis as generation,” Microsoft wrote, “this architecture creates a powerful feedback loop that delivers higher-quality results across factual accuracy, analytical breadth, and presentation.”
Critique mode will be automatically enabled when using Copilot’s Researcher agent, but users can also choose to have a single OpenAI or Anthropic model handle the entire research process.
How Anthropic's Claude AI Became a Co-Founder
With this new Critique mode, Microsoft said, Copilot Researcher now tops a Perplexity-developed benchmark that judges deep research models’ accuracy, completeness, and objectivity. Additional information on how Microsoft evaluated the new mode can be found here.
In addition, Microsoft announced a new “Council” mode for the Researcher agent that simultaneously uses an Anthropic model and an OpenAI model to complete the same task, side-by-side. “Once both reports are generated,” the company wrote, “a dedicated judge model evaluates the reports to create a distilled summary of key findings and highlights where the models meaningfully agree or diverge.”
Microsoft also announced said that Copilot Cowork, a feature based on Anthropic’s Claude Cowork tool. is now available for early adopters through the company’s Frontier program. The program enables users to access tools that are still in development and not quite ready for launch.