Microsoft Brings New AI Capabilities to Copilot Researcher

Share This Post

Microsoft released two new capabilities for Researcher, its AI research agent for enterprises within its Copilot platform.

The updates, dubbed Critique and Council, are designed to improve the agent’s accuracy and “depth,” as Microsoft seeks to bolster the reliability of AI-generated research for enterprises.

In a blog post on the launch on Monday, Microsoft said the features better an agent’s ability to handle complex research tasks, combining multiple AI models and using the same evaluation processes typically seen in academic and professional research. 

Critique works by dividing the research process between two AI models. One is used for planning, retrieval and drafting, while the second evaluates accuracy, strengthens arguments and refines the final report.

By using two AI workflows (rather than a single model), Critique is designed to boost source reliability and transparency.

In performance assessments, the system was found to surpass single-model approaches across categories including accuracy, analytical breadth and presentation quality, the vendor said.

Related:OpenAI Now Valued at $852B After Funding Round

“Critique pushes Researcher to identify missing analytical angles, close coverage gaps, sharpen formulations, and produce responses with stronger organization and clearer narrative flow,” Microsoft said in the blog. “This accounts for the substantial improvements to the breadth and depth and presentation quality scores.”

Council was also designed to improve reliability, offering a side-by-side comparison of outputs from multiple AI models.

When activated, Council runs models from Anthropic and OpenAI in parallel, with each generating a complete, independent report. A separate “judge” model then evaluates both outputs to produce a summary highlighting key areas of agreement and divergence, along with any particular points of note.

Microsoft said the feature is designed to give users greater transparency into how different AI systems approach the same problem — an increasingly important consideration as companies deploy multi-model systems.

Both models are now available in Microsoft’s Frontier program.

Related Posts

The SEC meeting that wasn’t: State of Crypto

Earlier this month, as it became clear that the...

How the EU’s new crypto rules triggered a massive scam wave

The European Securities and Markets Authority (ESMA) confirmed it...

Crypto investors are looking past market-cap rankings and back to fundamentals

Perpetual-futures volumes still run at a multiple of spot...

Safepal security vulnerability exposes data of 39,798 customers

The breach has impacted 39,798 customers who placed orders...

Ethereum Devs to Narrow 66 Proposals tied to 2027 Hegotá Upgrade

Ethereum developers are currently reviewing 66 proposals to narrow...

Crypto’s week in 5 stories

That may happen. But last week showed that right...