News
As Agentic AI Explodes, Microsoft Announces MS365 Copilot Agent Debugging
Microsoft announced agent debugging functionality for Microsoft 365 Copilot directly from the AI tool itself, no Visual Studio 2022 or Visual Studio Code needed.
Agentic AI is dominating AI these days, but it didn't allow for easy debugging. Debugging AI agents is challenging due to their unpredictable behavior, reliance on external inputs, and limited visibility into their decision-making processes. Issues can arise from integration failures, data misalignment, or model configurations, making it difficult to trace errors.
While debugging has generally been the province of IDEs and code editors, Microsoft's approach for debugging agents in Microsoft 365 uses the tool itself from within one of the softwares such as Word, Excel, Outlook and so on.
The process is somewhat akin to browser-based dev tools that are opened up with a command. All users have to do within one of the Microsoft 365 apps is type the -developer on command from the Copilot interface.
[Click on image for larger view.] -developer on (source: Microsoft).
"We're excited to announce the general availability of the agent debugging experience, now live in Microsoft 365 Copilot," said Microsoft's Carol Mbasinge Kigoonya in an April 9 post. "This milestone significantly enhances your debugging capabilities, observability, and troubleshooting for agents, empowering developers with improved tools to streamline their workflows."
Along with an introductory video, the announcement mostly consists of a bullet-point list of key features:
- Agent configuration insights
- Agent capabilities: Quickly identify active capabilities and their configurations.
- Knowledge scopes: Easily understand the data and resources accessible at runtime.
- Agent actions: View clearly detailed action configurations, function matching, and execution details, similar to previously supported plugin insights. This transparency helps ensure your agents perform as expected.
- Improved execution and performance monitoring
- Success & failure insights: Quickly identify and troubleshoot issues by monitoring capability and action outcomes.
- Execution details: Gain a better understanding of your agents' performance by tracking execution details and result count to optimize workflows.
- API plugin response details: Access important API plugin response details, including endpoints and headers (with redacted authentication tokens), to troubleshoot more effectively.
- Observability & troubleshooting tools
- Agent ID and conversation ID tracking: Easily track interactions using standardized identifiers (titleId.manifestId and Conversation ID), improving visibility and troubleshooting.
- Latency information: Identify and address performance bottlenecks quickly with detailed API plugin latency insights.
- Simplified collaboration: Share debugging details effortlessly using the Quick Copy Debugging JSON features.
More information is available in Debug an agent with developer mode documentation.
About the Author
David Ramel is an editor and writer at Converge 360.