Add post: OpenTelemetry agentic AI spec changes and Azure AI Foundry observability#19
Add post: OpenTelemetry agentic AI spec changes and Azure AI Foundry observability#19fusionet24 wants to merge 1 commit into
Conversation
…observability Covers the latest OpenTelemetry GenAI semantic conventions for multi-agent systems, the Microsoft/Cisco collaboration on new spans and attributes, Azure AI Foundry + Application Insights integration, and the telemetry challenges that remain for agentic AI observability. https://claude.ai/code/session_01VoHVD1nxXpDm4SQshywbZQ
There was a problem hiding this comment.
Pull request overview
Adds a new Quarto blog post explaining recent OpenTelemetry GenAI semantic convention updates for multi-agent systems and how Azure AI Foundry/Application Insights are adopting them for agent observability.
Changes:
- Introduces a new post covering why traditional telemetry struggles with agentic AI and what the GenAI spec adds for agents/tools/tasks.
- Summarizes Microsoft/Cisco Outshift contributions to multi-agent span/attribute proposals and their status in the spec.
- Describes Azure AI Foundry + Application Insights integration and remaining observability challenges (cost/content, evaluation, visualization, stability, sensitive data).
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| ::: {.callout-note appearance="simple"} | ||
| This post was developed with the help of AI based on my research. | ||
| ::: |
There was a problem hiding this comment.
AI-assisted posts elsewhere include a link to the AI Content Labels page in this same callout (e.g., posts/agentic-diagramming/index.qmd:16 and posts/agentic-decomposition/index.qmd:16). Consider adding that link here as well so readers can understand what “AI Assisted” means in your labeling scheme.
| | Tool Type | Description | | ||
| |-----------|-------------| | ||
| | `extension` | Agent-side tool calling external APIs directly | | ||
| | `function` | Client-side tool where the agent generates parameters and the client executes | | ||
| | `datastore` | Tool for accessing structured/unstructured data (RAG, knowledge bases) | | ||
|
|
There was a problem hiding this comment.
The tool-type table has an extra leading pipe on each row (e.g., || Tool Type | Description |), which will render as an unintended empty first column in Markdown/Quarto. Remove the extra leading | so the table renders correctly.
|
|
||
| This is where the spec meets implementation. [Azure AI Foundry](https://learn.microsoft.com/en-us/azure/ai-foundry/observability/how-to/trace-agent-setup) stores traces in Azure Application Insights using these OpenTelemetry semantic conventions. The integration is straightforward. Here's the core setup in Python: | ||
|
|
||
| ```python |
There was a problem hiding this comment.
This fenced block is marked as python but contains a pip install ... shell command. Consider changing the fence to bash/sh (or plain) so the syntax highlighting and copy/paste expectation matches the content.
| ```python | |
| ```bash |
| --- | ||
|
|
||
| ::: {.callout-note appearance="simple"} | ||
| This post was developed with the help of AI based on my research. |
There was a problem hiding this comment.
The callout text has a minor grammar issue: “AI based” should be hyphenated as “AI-based”.
| This post was developed with the help of AI based on my research. | |
| This post was developed with the help of AI-based on my research. |
Covers the latest OpenTelemetry GenAI semantic conventions for multi-agent
systems, the Microsoft/Cisco collaboration on new spans and attributes,
Azure AI Foundry + Application Insights integration, and the telemetry
challenges that remain for agentic AI observability.
https://claude.ai/code/session_01VoHVD1nxXpDm4SQshywbZQ