Microsoft Copilot Agents Fail at Everyday Business Tasks

Microsoft Copilot Analyst agents struggle with basic business tasks like writing memos and organizing meetings. Despite heavy investment in AI infrastructure and model licensing, these agents show limited competence compared to tools like GitHub Copilot. Ed Bott tested them over weeks on real workloads and found consistent errors and incoherence. The agents lack reliability for practical use, undermining claims of productivity gains in corporate environments. This gap highlights a disconnect between Microsoft's ambitious agentic OS vision and actual performance.