OpenTelemetry's GenAI semantic conventions define common names for model calls, agent and tool spans, MCP operations and token metrics. Coding agents such as Claude Code and Codex export OTel events for prompts, tool decisions and tool results, with prompt content redacted by default. Standardize on these names so detections work across vendors.
Read in this order
- ArticleAI Agent Observability: Evolving Standards and Best Practices (OpenTelemetry blog)Start here: why agent observability needs shared standards.
- OfficialSemantic conventions for GenAI agent and framework spansThe span names and attributes for agents and tool calls.
- VideoOpenTelemetry GenAI in Practice: What the Spec Says Vs. What You Actually See (CNCF [Cloud Native Computing Foundation])Compares real emitted traces with the semantic conventions. 24 min.