# @ai-sdk/workflow 2.0.94 adds single telemetry traces for durable agent runs

published: 2026-11-03T02:05:43.833Z
topic: frameworks
source: https://github.com/vercel/ai/releases/tag/%40ai-sdk%2Fworkflow%402.0.94
url: https://aiagentnews.io/stories/ai-sdk-workflow-2-0-94-adds-single-telemetry-traces-for-durable-agent-runs

Vercel's @ai-sdk/workflow 2.0.94 adds one telemetry operation covering durable WorkflowAgent runs, including prompts, final output, steps and tools. It also adds an optional OpenTelemetry bridge that carries trace context and sampling across workflow steps.

The patch records one telemetry operation for durable WorkflowAgent runs, covering the prompt, the final output, steps and tools. Events recorded inside the workflow are reported when a run completes or suspends, and they keep their original timestamps.

For OpenTelemetry, developers pass the same OpenTelemetryIdGenerator instance to the tracer provider's idGenerator and sampler, and to new OpenTelemetry({ durable }). The bridge carries parent trace context and sampling decisions across steps, reports model spans live with provider requests nested beneath them, and rebuilds the invoke_agent tree when the run finishes.

The release also adds typed experimental durable context hooks for telemetry integrations and drops reliance on process-global ID-generator detection. Dependencies move to @ai-sdk/provider-utils@5.0.74 and ai@7.0.162.

## Why it matters if you build agents

If you run durable agents with WorkflowAgent, you can get one telemetry tree per run instead of stitching separate spans together. Without the bridge, you keep the existing per-model live tracing.

Teams using OpenTelemetry sampling need to pass the same ID generator and sampler to both the tracer provider and the bridge. The new hooks are marked experimental, so pin the version before building on them.