AI Foundry, copilot-studio

What’s New in Copilot Studio: July 2025 – Advancing Agentic Capabilities

The primary focus of this update was the introduction of Agentic Actions and Dynamic Chaining, shifting the platform from simple conversation-based triggers to reasoning-based autonomous execution.

Real-time User Journey: Dynamic Chaining

The core “journey” updated in July 2025 involves how an agent handles a multi-step request without manual “if/then” branching:

  1. Complex Prompt: A user asks, “Find the latest sales report for the Northeast region, summarize the top three concerns, and email that summary to the Regional Manager.”
  2. Reasoning: Instead of looking for a specific pre-built “topic,” the agent uses its Generative AI orchestrator to identify the necessary steps.
  3. Dynamic Action: The agent autonomously “chains” together three separate tools:
    • Tool 1 (SharePoint): Searches for and retrieves the specific PDF/Excel file.
    • Tool 2 (LLM Reasoning): Analyzes the content of that file to extract the top three concerns.
    • Tool 3 (Outlook): Drafts and sends the email to the specific contact.
  4. Completion: The agent confirms the email has been sent and provides a copy of the summary to the user in the chat.

Step-by-Step: How to Enable Dynamic Chaining

To move your agent from “Classic” mode to the new “Agentic” reasoning mode:

  • Step 1: Open Agent Settings: In Microsoft Copilot Studio, select your agent and go to the Generative AI page in the side navigation.
  • Step 2: Switch Orchestration: Under the “Orchestration” section, select Dynamic Chaining (Preview). This allows the agent to use generative AI to choose the best topic or action to fulfill a request.
  • Step 3: Add Actions: Navigate to the Actions tab. Add the specific “skills” the agent can use (e.g., pre-built connectors for Outlook, SharePoint, or custom Power Automate flows).
  • Step 4: Refine Descriptions: For every action, provide a clear, natural language Description. This is critical, as the agent uses these descriptions to “know” when to trigger that specific tool.
  • Step 5: Test & Publish: Use the Test Pane to verify the agent can correctly chain these tools together before publishing to your users.

Infographic: The Shift to Agentic AI

The July 2025 update defined a clear evolution of the platform’s architecture:

FeatureTraditional Bot (Legacy)Agentic AI (New)
Logic EngineFixed Topic Triggering (Keywords)Generative AI Orchestration
WorkflowLinear/Pre-defined PathsDynamic Chaining (Reasoning-based)
Tool IntegrationManual API CallsAutonomous Tool Use via Descriptions
User InputsSpecific Phrases RequiredComplex, Multi-intent Prompts

References

Leave a comment