Microsoft on Monday announced enhancements to its Copilot research assistant that enable simultaneous use of different AI models inside a single workflow, part of the company's push to refine its AI tools and accelerate customer adoption.
The centerpiece of the update is a feature called "Critique." Under this workflow, Copilot's Researcher agent will produce outputs using both OpenAI's GPT and Anthropic's Claude models for each response rather than relying on only one model. In the current setup, GPT generates the draft response while Claude evaluates that output for accuracy and quality before it is returned to the user, Microsoft said. The company indicated plans to make the review process bi-directional in the future, so GPT could likewise critique drafts produced by Claude.
Nicole Herskowitz, corporate vice president of Microsoft 365 and Copilot, described the capability as a way to capture the benefits of having different vendor models in a single assistant while enabling those models to work together rather than operating in isolation. She said the multi-model approach is intended to speed user workflows, help control AI hallucinations - where systems produce incorrect or fabricated material - and deliver more dependable outputs to customers, improving both productivity and quality.
Alongside Critique, Microsoft is introducing a tool called "Model Council" that allows users to view and compare responses from several AI models side-by-side. This feature is designed to give customers clearer visibility into differences in model behavior and output, according to the company.
Microsoft is also broadening access to Copilot Cowork, an agentic AI tool, by making it available to more members of its Frontier program, which offers early access to select new AI features. Copilot Cowork was first shown in testing mode earlier this month and is based on Anthropic's Claude Cowork product, tapping into market interest in autonomous AI agents.
The company has been working to refine Copilot as it faces competition from other advanced AI systems such as Google's Gemini and external autonomous agents like Claude Cowork. Microsoft said the updates are intended to enhance the utility and reliability of Copilot for customers as the ecosystem of models and agentic tools expands.
In addition to product details, the announcement included a reference to Microsoft stock valuation tools used to assess whether Microsoft is a bargain at current prices. The text noted a Fair Value calculator that applies a group of valuation models to provide a view on MSFT's valuation and to help identify other investment opportunities.
Summary
Microsoft unveiled a "Critique" workflow for Copilot that runs GPT and Claude together - with GPT drafting and Claude reviewing - plus a Model Council comparison tool. The company also widened early access to Copilot Cowork through its Frontier program; Copilot Cowork is based on Anthropic's Claude Cowork and was previously in testing.
Key points
- Copilot's new Critique feature uses both GPT and Claude on every response, with Claude reviewing GPT's output for accuracy and quality.
- Model Council will let users compare outputs from different AI models side-by-side, increasing transparency in model behavior.
- Copilot Cowork, an agentic tool derived from Anthropic's Claude Cowork, is being opened to more Frontier program members, expanding early access to Microsoft's autonomous agent capabilities.
Risks and uncertainties
- Effectiveness risk: The updates aim to reduce AI hallucinations, but the company did not provide empirical performance metrics in this announcement; outcomes for accuracy and reliability remain to be demonstrated - impacting enterprise software and cloud services users.
- Adoption risk: Bringing multiple models together and expanding agent access increases complexity for customers, which could affect uptake among business and developer users in software and cloud markets.
- Competitive pressure: Microsoft is advancing Copilot amid competition from other AI offerings and autonomous agents; relative market acceptance is uncertain and could affect technology and cloud infrastructure segments.