Microsoft has introduced early access to Copilot Cowork through its Frontier program, marking a significant step in the company’s push to move artificial intelligence beyond passive assistance and into the territory of actively executing complex work on behalf of users. The rollout arrives alongside a set of meaningful updates to the Researcher tool, including a new Critique feature and a Model Council capability, all forming part of Wave 3 of Microsoft 365 Copilot, which is centred on enabling artificial intelligence to perform substantive work tasks rather than simply supporting them.
Copilot Cowork is an agent-based artificial intelligence system designed to handle multi-step tasks within the Microsoft 365 environment, built on the foundation of Claude Cowork developed by Anthropic. The tool operates by allowing users to define a desired outcome, after which it constructs a structured execution plan and carries out the task while displaying real-time progress. Users retain the ability to intervene at any point to adjust or redirect the workflow, making the system collaborative rather than fully autonomous. It supports both standalone tasks and recurring workflows such as monthly budget reviews, and operates across documents and work environments in a manner that functions much like a capable digital collaborator working alongside the user rather than waiting to be prompted at each step.
On the research side, Microsoft has added a Critique feature to its Researcher tool that introduces a dual-model review process. In this setup, GPT-4 generates the initial response to a query while Claude, developed by Anthropic, independently reviews the output for accuracy and quality before it is surfaced to the user. According to Reuters, Microsoft is also exploring expanding this into a fully bidirectional process in which GPT-4 could similarly review Claude-generated responses in future iterations. Microsoft reported that the Critique feature improved the Researcher tool’s performance by 13.8% on the DRACO benchmark, a measure of deep research accuracy and output quality. Complementing this is the Model Council addition, which allows users to simultaneously generate responses from multiple artificial intelligence models and compare them side by side, with the interface highlighting differences, similarities, and distinct contributions from each model to give users greater transparency and informed control over the artificial intelligence outputs they choose to act upon.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
