News

Microsoft Gives Orgs More Power to 'Tune' AI Agents

At its Build 2025 conference this week, Microsoft unveiled significant advancements aimed at empowering enterprises to create more sophisticated AI agents.

Central to these developments are two key initiatives: Copilot Tuning and the integration of the Model Context Protocol (MCP), both designed to enhance the customization and interoperability of AI agents within enterprise environments.

Copilot Tuning enables organizations to fine-tune Microsoft's AI copilots using their proprietary data, allowing for the creation of agents that align closely with specific business processes and communication styles. This feature facilitates the development of AI agents that can operate with a deeper understanding of organizational context, improving their effectiveness in tasks such as drafting communications, analyzing data, and automating workflows.

At the same time, Microsoft's support for MCP -- a protocol introduced by Anthropic -- aims to standardize the way AI agents interact with various tools and data sources, promoting seamless collaboration between agents from different vendors and enhancing their ability to perform complex, multi-step tasks autonomously.

Copilot Tuning Brings Low-Code AI Customization to the Enterprise
Microsoft 365 Copilot Tuning, announced on Monday, is a low-code solution that allows organizations to train AI models using their own data, workflows and processes

With Copilot Tuning, companies can build specialized agents tailored to domain-specific tasks. For example, a legal firm can train an agent to draft documents in its preferred language and tone, while a consulting firm can fine-tune agents for industry-specific expertise. These agents operate within the secure Microsoft 365 boundary, and Microsoft says customer data is not used to train foundation models.

Copilot Tuning will be available starting in June through an early adopter program.

Multi-Agent Orchestration and Model Flexibility in Copilot Studio
Already used by more than 230,000 organizations, Copilot Studio now supports multi-agent orchestration, allowing agents to collaborate and divide tasks based on their areas of expertise. In public preview, this capability enables scenarios such as automated employee onboarding -- where HR, IT and operations agents coordinate in parallel to streamline the process.

Microsoft also announced broader model interoperability in Copilot Studio through integration with Azure AI Foundry, giving developers access to over 1,900 models, including industry-specific LLMs. This “bring-your-own-model” feature helps align agent behavior with business-specific terminology and logic.

Expanded Developer Tools for Agent-Centric Workflows
Microsoft is investing heavily in enabling pro-code development alongside low-code tools. The new Microsoft 365 Agents Toolkit, now generally available, includes a software development kit for debugging and deploying agents across Microsoft 365 apps and Teams.

Meanwhile, a new Teams AI library aims to help developers optimize agents for chats, channels and meetings, with support for open standards such as the Agent-to-Agent (A2A) protocol and MCP. Microsoft also previewed Microsoft 365 Copilot APIs, starting with retrieval and chat capabilities, to let developers embed Copilot features into custom applications while respecting organizational permissions.

Developers can now oversee and manage agent workflows using a new Agent Feed within Power Apps, while Solution Workspace streamlines the app-building process with generative UI capabilities and support for code-first deployment in Visual Studio Code.

Windows 11 Secures the Model Context Protocol for Agent Interoperability
On the OS front, Microsoft detailed how Windows 11 will adopt and secure the Model Context Protocol (MCP) -- an emerging standard for inter-agent communication and orchestration.

MCP allows agents and applications to discover and invoke tools in a consistent, HTTP-based format, but its openness introduces new risks, including prompt injection, command injection and tool poisoning.

To counter these threats, Microsoft is building a secure MCP architecture into Windows 11 that includes:

  • Proxy-mediated communication, where all MCP traffic is routed through a trusted Windows component that enforces centralized policies and auditing.
  • Tool-level authorization, requiring user approval for each agent-tool interaction.
  • A central registry of vetted MCP servers meeting baseline security requirements.
  • Runtime isolation and privilege enforcement to minimize impact in case of compromise.

MCP servers listed in the Windows registry must meet strict criteria, including mandatory code signing, immutable tool definitions, declared privilege scopes and security-tested interfaces. An early developer preview of the MCP platform will roll out after Build 2025, with secure-by-default enforcement expected in the coming months.

About the Author

Chris Paoli (@ChrisPaoli5) is the associate editor for Converge360.

Featured

  • IBM Giving Orgs a Governance Lifeline in Agentic AI Era

    Nearly overnight, organizations are facing brand-new challenges caused by self-directed AI systems (a.k.a. agentic AI). Big Blue is extending them some help.

  • Microsoft Launches Integrated E-mail Security Ecosystem for Defender for Office 365

    Microsoft is expanding its e-mail security capabilities with the launch of a new Integrated Cloud Email Security (ICES) ecosystem for Microsoft Defender for Office 365.

  • Microsoft Joins Workday's AI Agent Partner Network

    Microsoft has become a key partner in Workday's newly launched AI Agent Partner Network, aligning with other industry leaders to integrate AI agents into enterprise workforce systems.

  • LinkedIn CEO Ryan Roslansky To Lead Microsoft's Productivity Initiatives

    In a strategic leadership realignment, Microsoft has appointed LinkedIn CEO Ryan Roslansky to oversee its consumer and small business productivity software division, encompassing Microsoft 365, Teams and AI-driven tools like Copilot.