I paid Microsoft's premium Copilot agents to do my work - they were confidently bad at it (www.zdnet.com)

A recent experiment with Microsoft's Copilot agents revealed significant shortcomings in their ability to assist with everyday work tasks, despite the company's heavy investment in AI technology. While Microsoft aims to develop a suite of "agentic" tools for automating routine corporate tasks, users are finding that the Copilot features fall short, often delivering a mix of irrelevant information and ineffectual suggestions. The tested agents occasionally demonstrated competence, yet failed in practical execution—for instance, a request for a functional Excel file resulted in a broken link, underscoring their limitations. This investigation is significant for the AI/ML community as it highlights the gap between ambitious goals of AI systems and their actual performance in real-world applications. Despite being underpinned by substantial resources, including partnerships with leading AI firms like OpenAI and Anthropic, the Copilot agents struggle with basic tasks, raising questions about the effectiveness of current AI training and deployment strategies. As enthusiasm for AI tools grows, experiences like this could serve as crucial feedback for improving AI capabilities and enhancing user trust in these technologies.