OpenTelemetry Semantic Conventions for AI Agent Observability: Standardized Span Formats for Tasks, Actions, Agent Teams, Artifacts, and Memory Operations

OpenTelemetry·medium signal

OpenTelemetry merged semantic conventions defining standard attribute names and span formats for AI agent systems. Conventions cover tasks (minimal trackable work units decomposable into subtasks), actions (tool calls, LLM queries, API requests, vector DB queries, human input), agent teams, artifacts, and memory operations. MCP tool executions trace via corresponding MCP instrumentation. OpenLLMetry ships an MCP server bridging production telemetry into dev tooling. Integrates with Grafana, Datadog, Honeycomb, Splunk, and New Relic using standard attribute names. Enables cross-framework observability across CrewAI, AutoGen, LangGraph.