Microsoft has officially unveiled a major update for its artificial intelligence assistant, introducing new features as part of the Microsoft 365 Copilot Wave 3 release. The latest upgrade brings advanced capabilities to the workplace, including a task automation tool called Copilot Cowork and enhanced multi-model research functions. These tools are designed to shift artificial intelligence from handling simple, single prompts to managing complex, multi-step workflows. Currently, these new capabilities are available to early access customers through the company’s Frontier program.
Copilot Cowork Transforms Task Delegation
The standout addition to the update is Copilot Cowork. Built upon technology from Anthropic’s Claude Cowork, this feature allows users to delegate ongoing, multi-step projects rather than just asking for quick answers. Users simply describe their desired outcome, and the system autonomously develops a plan. It then navigates through workplace tools and files to execute the necessary tasks.
Throughout the process, the assistant displays its progress transparently, allowing workers to intervene and guide the direction of the project at any time. The system includes built-in skills from both Microsoft and Claude. These integrated plugins help automate specific business activities, ranging from managing calendars and generating daily task briefings to handling repeatable processes like monthly budget reviews.
Jared Spataro, Microsoft’s chief marketing officer for AI at Work, emphasized that this tool makes task delegation effortless. According to Spataro, the system reasons across enterprise tools to carry work forward while providing opportunities for human steering.
Early testing has already shown significant productivity gains for enterprise users. Barton Warner, a senior vice president of enterprise technology at Capital Group, shared that his firm has used the system since earlier this year. He noted that the new capabilities allow them to scale their artificial intelligence ecosystem securely. Warner highlighted that the tool is not merely generating text; it takes real action by coordinating tasks across everyday workflows. Because the system operates strictly within the firm’s security boundaries and relies on internal enterprise data, teams can experiment and scale their operations with confidence.
Smarter Research With the Critique Function
Beyond task automation, Microsoft has significantly upgraded its internal research tool by moving away from relying on a single artificial intelligence model. The revamped Researcher feature now employs a multi-model approach, utilizing technology from frontier labs like OpenAI and Anthropic.
The core of this upgrade is a new function called Critique. Instead of one model doing all the work, multiple agents collaborate and hand off tasks to one another to ensure maximum quality. In practice, a GPT model plans the research and generates an initial draft. Following this, a Claude model steps in as an expert reviewer. This second agent scrutinizes and refines the draft to guarantee accuracy, objectivity, and completeness before presenting the final response to the user.
Microsoft CEO Satya Nadella highlighted the significance of this development, noting that using multiple models together generates optimal reports and responses. He stated that benchmark testing proves the system delivers industry-leading deep research capabilities.
Performance metrics support these claims. The updated Researcher tool scores heavily on the DRACO benchmark, which measures Deep Research Accuracy, Completeness, and Objectivity. The multi-model setup achieved a score of 57.4 percent, outperforming single-model systems by 13.8 percent. According to early testing data, this configuration is more than twice as reliable as standalone setups like OpenAI’s o4-mini model. It also scored higher than several other competitors, including Gemini Deep Research, Claude Opus 4.6, and Perplexity’s Deep Research. However, the company did not publish direct comparisons against newer, flagship single models like GPT-5.4.
Reflecting on these advancements, Spataro stated that when trust and intelligence align, artificial intelligence transitions from an experimental phase into the standard way work is accomplished. He defined this Wave 3 progress as an intelligence that truly understands the context of the workplace.
Diverse Perspectives With Model Council
To further leverage the power of different artificial intelligence systems, the update introduces a feature known as Model Council. This tool allows workers to submit a single question and simultaneously view how different models respond to it.
By comparing the outputs side-by-side, users can immediately identify where the different systems agree and where their perspectives diverge. It also highlights the unique insights each model brings to the table. Microsoft compares this experience to having a team of different human researchers available all at once, offering diverse viewpoints on complex issues.
Availability and Future Rollout
While the Wave 3 features are currently limited to early access participants in the Frontier program, a broader release is planned for the future. Once the testing phase concludes and the system is fully launched, Copilot Cowork and the advanced research tools will be accessible to enterprise customers subscribed to the Microsoft 365 E7 AI tier.
