Observe AI SDK Agent

Wire production visibility into an existing TypeScript or JavaScript AI SDK agent. Prefer the SDK's first-party telemetry, fill only the gaps the product needs, and close the loop by turning sanitized failure traces into eval fixtures for build-ai-sdk-agent-evals and baseline evidence for optimize-ai-sdk-agent.

baseline -> enable native telemetry -> instrument gaps -> attribute cost ->
export reliably -> capture safely -> bridge to evals -> alert (detect only)

Scope and hard guards

Work on an existing AI SDK agent or agent-backed chat feature. Add or change instrumentation, exporters, sampling, redaction, alert definitions, and the smallest production seams needed for identity propagation and failure-trace export. Do not redesign the agent, rewrite prompts for quality, or implement spend enforcement.

Keep these guards active throughout the run: