Does OTEL GenAI cover everything I need for agent observability?

Not yet. It covers LLM calls, tool calls, and agent spans well, but eval scores, human-labeled quality signals, and reasoning-trace bodies are still vendor-extension fields. Assume you'll need some platform-specific configuration for those; the base traces are portable.

OpenTelemetry GenAI

The OpenTelemetry semantic conventions for AI/LLM workloads — a standardized set of span attributes (`gen_ai.request.model`, `gen_ai.usage.input_tokens`, `gen_ai.tool.name`, etc.) that agent observability platforms emit and ingest, replacing 2023-era vendor-proprietary trace formats. Matured through 2025 to the point that most 2026 platforms (Langfuse, Arize, LangSmith, Braintrust, Datadog, Grafana) support OTEL GenAI either natively or via ingest adapters. Instrumentation SDKs — OpenLLMetry (Traceloop), OpenInference (Arize), and LangChain's OTEL integration — emit conforming spans so you can change backends without re-instrumenting.

Written by Max Zeshut

Founder at Agentmelt · Last updated Sep 9, 2026

Frequently asked questions

Does OTEL GenAI cover everything I need for agent observability?: Not yet. It covers LLM calls, tool calls, and agent spans well, but eval scores, human-labeled quality signals, and reasoning-trace bodies are still vendor-extension fields. Assume you'll need some platform-specific configuration for those; the base traces are portable.

The pattern behind it

Evaluation & learning loops

Explained in plain words, with every automation that uses it.

Or get a plan for your own process

Three sentences; one option and its price appear in minutes. No call.

Free audit