Microsoft announced significant upgrades to its 365 Copilot platform, including enhancements to its Researcher agent and the soft launch of the Copilot Cowork feature. The updates were detailed in a blog post released on March 30.
The new features aim to improve AI-driven research capabilities, impacting how users generate and utilize reports. Microsoft stated that the upgrades enhance the efficiency and quality of research processes for 365 Copilot users.
One of the key additions is a “Critique” mode for the Researcher agent, which utilizes two distinct models: one for report generation and another as an expert reviewer. Microsoft emphasized that this dual approach facilitates a feedback loop that improves factual accuracy, analytical depth, and presentation quality. According to Microsoft, this mode is automatically enabled for users but allows them the option to revert to a single model from OpenAI or Anthropic.
With the implementation of Critique mode, Microsoft claimed that Copilot Researcher has surpassed a benchmark developed by Perplexity, which assesses deep research models on accuracy, completeness, and objectivity.
Additionally, Microsoft introduced a “Council” mode that operates both an Anthropic model and an OpenAI model simultaneously for the same task. Once both reports are produced, a judge model evaluates them, generating a distilled summary of key findings while highlighting areas of consensus or divergence.
Microsoft also announced the availability of the Copilot Cowork feature, which is based on Anthropic’s Claude Cowork tool. This feature is currently accessible to early adopters through Microsoft’s Frontier program, designed for users to test tools still in development.
