Microsoft Expands Copilot Cowork to Frontier Program, Upgrades Researcher With Multi-Model Intelligence

You’re reading an issue of "The AI Economy," my newsletter exploring the forces shaping the AI era—tracking how AI is rewriting business, work, technology, and culture. Subscribe to get expert insights and curated updates delivered straight to your inbox.

Microsoft is opening Copilot Cowork to a broader set of enterprise customers through its Frontier program, the company announced today—the latest step in its bet that AI should stop answering questions and start finishing work. Powered by Anthropic’s Claude, Copilot Cowork lets users delegate complex tasks to an AI agent that plans and executes work independently, checking in when it needs direction. Microsoft is also releasing an upgraded Researcher agent equipped with what it calls “multi-model intelligence.”

As organizations continue to transform into “Frontier Firms,” work is no longer restricted to a single LLM. Instead, Microsoft is pitching a reality in which multiple AI engines are utilized to power a single task. Copilot Cowork is its most tangible proof of that vision so far. Users describe their intended outcome, and Copilot Cowork will draw up a plan, reason across their tools and files, and execute the necessary steps. It’s grounded in Microsoft’s Work IQ and protected by Enterprise Data Protection.

Greater Access to Copilot Cowork

Since its debut at the beginning of March, Copilot Cowork has been restricted to a limited set of customers in Microsoft’s Research Preview. It’s now accessible to those enrolled in the Frontier program, a beta channel for Copilot and AI agent testing.

“This isn’t about generating content or answers. It’s about taking real action—connecting steps, coordinating tasks, and following through across everyday workflows,” Barton Warner, the senior vice president of enterprise technology at Capital Group, says in a statement. “Because Cowork operates on our enterprise data and within our security and risk boundaries, we can experiment, learn, and scale with confidence. This allows us to move faster and focus AI in places where it actually delivers value.”

M365’s Researcher Agent Goes Multi-Model

Further demonstrating its embrace of multi-model intelligence, Microsoft is upgrading its Researcher agent. Introduced in Microsoft 365 Copilot a year ago, this agent uses OpenAI’s deep research model and Microsoft’s tech to help workers with complex, multi-step research. Now, Researcher includes a Critique feature that puts GPT and Claude to work together to generate a response.

“GPT drafts, Claude reviews for accuracy, completeness, and citation integrity before it’s delivered,” Jared Spataro, Microsoft’s chief marketing officer for its AI at Work division, writes. In other words, Claude becomes the auditor of GPT’s work, reviewing every response before it reaches the user. He adds that at some point, “We expect this workflow to be bidirectional.”

The impact of this could be profound, as it offers workers peace of mind, knowing that some verification is taking place. Put differently, it’s answering the question, “Who watches the watchmen?” Workers can use the Researcher agent for a variety of use cases, from building detailed go-to-market strategies to conducting market research, competitor analysis, or preparing for meetings. And because it can connect to data from Salesforce, ServiceNow, and Confluence, it’s critical that the work it generates is free of hallucinations and as accurate as possible. The Critique feature is a built-in check on the quality of AI-generated work—trust, but verify.

Microsoft touts that early evaluations show that Researcher with Critique outperforms Perplexity Deep Research, Claude Opus, Gemini Deep Research, and OpenAI Deep Research on the Deep Research Accuracy, Completeness, and Objectivity benchmark. It’s an industry standard for measuring research quality.

One other update coming to Researcher is the addition of Microsoft’s Council model. Workers can use it to evaluate responses from different models in a side-by-side comparison. Specifically, it shows where models agree and disagree, as well as what each LLM brings to the table.

Together, these updates are part of Wave 3 of Microsoft 365 Copilot. Microsoft frames it as an effort to pair AI intelligence with enterprise trust, moving Copilot from a tool workers try to one they rely on. While the vision is interesting, will it help Microsoft drive commercial sales growth? In January, Chief Executive Satya Nadella revealed that there were 15 million paid M365 seats, accounting for three percent of commercial M365 subscriptions.

Featured Image: Satya Nadella announces at Build 2025 the general availability of Microsoft 365 Copilot. Credit: Ken Yeung