QuestionQ93

VerifiedImplement generative AI and agentic solutions

You have a Microsoft Foundry project that contains a multi-agent solution. The agents use tool calling to query internal systems.

You need to implement responsible AI auditing to meet the following requirements:

  • Capture all the nested operations across the entire agent run.
  • Record tool invocation arguments and returned results as metadata.

What should you use for each requirement? Each option may be used once, more than once, or not at all.

Drag & Drop
Capture all the nested operations across the entire agent run:
Record tool invocation arguments and results:
Explanation

In Microsoft Foundry's OpenTelemetry-based agent tracing, a trace consists of spans that can be nested as parent/child spans, forming a hierarchical tree that reflects the full call stack of an agent run — including sub-agent calls and tool invocations. This hierarchical span structure is what allows capture of all nested operations across an entire agent run, and is rendered as a hierarchical trajectory/tree in the Foundry portal's Observability > Traces view. Separately, each tool-call span carries attributes (metadata) that record the tool's invocation arguments and returned results, following OpenTelemetry semantic conventions for tool calls, which makes this data queryable and auditable. KQL query filters and sampling/trace sampling policies relate to querying stored telemetry or controlling which traces are collected, not to the structural capture of nested operations or the recording of tool call inputs/outputs.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!