LangSmith Added End-to-End OpenTelemetry Support for LangChain and LangGraph

LangSmith — a platform for debugging and monitoring LLM applications from the LangChain team — has announced full OpenTelemetry support. Developers can now connect AI component tracing to a unified observability stack along with the rest of the application services, without maintaining two isolated monitoring tools in parallel.

Why AI services need OTel

OpenTelemetry (OTel) is an open CNCF standard for collecting distributed traces, metrics, and logs. It is supported by virtually all major monitoring platforms: Datadog, Grafana, Honeycomb, Jaeger, AWS X-Ray, New Relic. In production systems, OTel has become a de-facto requirement: it allows you to trace a user request's path through dozens of microservices, databases, and external APIs in a single temporal context.

Until now, AI components lived in a separate bubble. LangSmith stored LLM call traces in a proprietary format, while the rest of the infrastructure team used an OTel backend. During each production incident, engineers had to switch between two systems, manually matching timestamps and call context. This slowed down diagnostics and created blind spots in observability.

What changed in LangSmith

The new OTel support works bidirectionally and covers the entire LangChain ecosystem stack:

• Trace ingestion: LangSmith accepts data through the standard OTel collector without custom SDKs or special adapters
• Data export: LLM calls and agent traces can be sent to any OTel-compatible backend — Datadog, Grafana Tempo, Honeycomb, Jaeger
• End-to-end context: distributed traces are linked across service boundaries — from the user's incoming HTTP request to the language model response and back
• LangGraph support: in addition to linear LangChain chains, OTel tracing covers complex graph agents with cycles, parallel steps, and conditional branches

Key change for engineers: no longer need a separate LangChain SDK for instrumenting the AI part. The standard OTel agent, already configured for the backend, automatically begins collecting data from LangSmith.

What production teams gain

Imagine a typical incident: a user complains that a chatbot response takes 10-15 seconds. Previously, an engineer would open LangSmith, study the language model call trace, then switch to Datadog to check network delays, queues, and database — and manually stitch together the picture. Now the entire request path — from the frontend through the API gateway and orchestrator to the LLM call and final response processing — is visible in a single continuous trace with millisecond precision.

For teams moving prototypes into production load, this means standard OTel practices — alerting on p95/p99 latency, SLO metrics, error budgets — are now applicable to AI components without additional configuration.

"End-to-end observability is not just convenience, it is a mandatory

requirement for any production AI system serving real users under real load."

For enterprise customers, it is particularly important that corporate IT departments have already invested in OTel infrastructure and have strict standardization requirements. A new tool that requires its own isolated dashboard is always an additional barrier to adoption agreement. LangSmith eliminates this barrier.

What this means

LangSmith is moving toward standardizing AI observability. Instead of a proprietary approach of "everything only in our UI" — an open standard that fits organically into the team's existing tools. This makes monitoring of AI systems part of the normal engineering process, not a separate discipline with its own data silo. For the market overall — another step toward AI infrastructure being perceived as a standard part of the production stack, rather than an exotic addition.