At Build 2025, Microsoft announced Microsoft 365 Copilot Tuning, a new way for organizations to use their company data to train models that can perform better on domain-specific tasks. Models created using Microsoft 365 Copilot Tuning can be used by the agents built with Microsoft 365 Copilot’s Agent Builder.
For example, a legal firm can use the tuned model to create documents that take advantage of client-specific context. Another example is that a consulting company involved in the aviation industry can make an agent that takes advantage of a model tuned on documents dealing with specific regional or international regulatory requirements.
The Microsoft 365 Copilot Tuning Early Adopter Program will be available in June for customers with 5,000 or more Microsoft 365 Copilot licenses.
Microsoft also announced the public preview of multi-agent orchestration in Copilot Studio. With this capability, agents created in Copilot Studio can exchange data, collaborate on tasks, and take on tasks based on their expertise. For example, several agents can collaborate on various required tasks across HR, IT, and marketing to help onboard a new employee.
Next is the capability to bring models to Copilot Studio with Azure AI Foundry Models integration. With this integration, customers will have access to more than 1,900 models.
Finally, Microsoft is making it easier for organizations to manage agents using Microsoft Entra. Every agent created either through Copilot Studio or Azure AI Foundry will be assigned a Microsoft Entra Agent ID to give IT admins visibility and control. Microsoft is also expanding Microsoft Purview Information Protection to Copilot Studio agents that use Microsoft Dataverse. Both of these capabilities are now in public preview.
By enabling fine-tuning with proprietary data and multi-agent collaboration, Microsoft is paving the way for custom AI solutions across various industries.
0 Comments - Add comment