Microsoft Copilot Studio is giving makers new ways to build agents that reason through complex work, draw on richer context, and use a growing set of tools.

GitHub Copilot harness reaches general availability

The GitHub Copilot harness — a new foundation for reasoning-heavy agents and workflows — is now generally available. It sits between the AI model and the agent, orchestrating when to call the model, what context to provide, and which tools, MCP servers, or connected agents to use, letting agents adapt their approach as they work toward a goal instead of following a predefined path.

flowchart LR
    A[Goal defined] --> B[GitHub Copilot harness]
    B --> C[Choose tools, MCP servers, connected agents]
    C --> D[Plan and adapt steps]
    D --> E[Complete multistep business process]

Expanding what agents can do

  • Skills: package instructions into modular, reusable capabilities that can be added to multiple agents or shared with a team.
  • Connected agents (preview): let a primary agent delegate requests to specialized agents, enabling modular architectures.
  • Workflows and MCP servers as tools (GA): give agents access to deterministic multistep automation and external services.
  • Windows 365 for Agents MCP server (GA): lets agents interact with a Windows 365 Cloud PC to work with desktop apps, browsers, and UIs.

Richer context: knowledge, memory, and files

  • Work IQ (preview) connects agents to organizational context — emails, calendar, files, Teams messages, and people information.
  • Foundry IQ connections let agents ground responses in a knowledge base already built in Microsoft Foundry.
  • Memory (preview) lets agents retain preferences and patterns from previous interactions with a user.
  • Agents can now analyze uploaded files and users can view files an agent creates during a conversation.

Stronger governance

  • Since July 2026, Copilot Studio automatically creates a Microsoft Entra Agent ID for every new agent.
  • MCP servers can be submitted for Microsoft certification covering reliability, security, compliance, and responsible operation.
  • Environment-level agent telemetry (preview) can be exported to Application Insights via the Power Platform admin center.

Real-time agents and more model choice

Real-time agents can now be deployed to digital messaging channels in addition to voice (preview), with GPT-5 Chat available for voice experiences and Application Insights monitoring via Azure Monitor. Claude Sonnet 5 and GPT-5.5 Chat are now generally available as primary models alongside multiple harness options.