Microsoft Copilot Researcher Agent Just Got Smarter with Critique and Counsel!

Copilot Researcher just got a major upgrade aimed at fixing the biggest problem with AI research: confident answers that miss key context. Microsoft added two new features inside the Researcher agent that bring multi-model review into the workflow before you ever trust the output. In this video, you’ll learn what Researcher is and why it’s different from regular Copilot chat, then we break down the two upgrades. Critique adds a second model that reviews the first model’s research for source quality, completeness, and evidence grounding, then improves the final report. Counsel runs two models in parallel, compares where they agree and disagree, and surfaces unique insights each one found so you can triangulate high-stakes decisions. You’ll also learn the biggest limitation most people hit: Researcher has a monthly usage cap, what counts against it, what doesn’t, and how to be strategic so you don’t burn your 25 deep-research runs on questions that regular Copilot could handle. 00:00 Why AI research fails and what Microsoft just changed 00:51 What Researcher is and why it’s not regular Copilot chat 01:32 Feature 1 Critique, two-model peer review 02:31 What Critique checks source reliability completeness grounding 03:31 Critique is now the default Researcher behavior 03:48 Demo setup inside the Researcher agent 05:18 Auto mode uses GPT with Claude refinements 07:01 Feature 2 Counsel, parallel reports plus a judge summary 08:20 Counsel demo, compare GPT vs Claude and key disagreements 10:27 The limit 25 Researcher and Analyst runs per month and how to manage it