Agent bandwidth savings and AI tracing converge

Agent bandwidth gets a compact upgrade

The biggest operational change this week is the Datadog Agent 7.81.0 defaulting to the v3 metrics intake. The new payload format is more compact, so every Agent that updates will immediately start sending less outbound bandwidth to Datadog — a silent but meaningful win for teams running many Agents, especially in high-cardinality environments. If you're still relying on the v2 intake (for example, because you route through Observability Pipelines Worker), that path is unchanged. The release also renames a systemd unit from datadog-agent-procmgrd to datadog-agent-procmgr on Linux, and replaces the DDOT metric remapping feature gate — so a quick search for datadog-agent-procmgrd.service in your automation scripts is the wise move.

AI tracing fills out the stack

AI agent observability had a notably busy week. Datadog's JavaScript APM libraries — both dd-trace-js v6.2.0 and v5.113.0 — added support for the Mercurius GraphQL fastify plugin, producing a top-level graphql.request span. More significantly, Datadog APM v6.1.0 and v5.112.0 now trace Anthropic's Claude Agent SDK and Vercel AI SDK v7. On the Python side, the v4.12.0rc1 release candidate adds AI Guard standalone mode, meaning you can get AI Guard trace decisions without incurring APM host billing, and a new Ray Serve instrumentation for versions 2.47.1+. The v4.11.0 stable release also brings LLM span export and AI Guard for Anthropic, plus Data Streams Monitoring for Google Cloud Pub/Sub. Sentry's SDKs joined the party with Sentry JavaScript 10.65.0 adding URL attributes across a dozen frameworks and native instrumentation for MySQL2 and GraphQL v17. A recurring theme: Datadog's Python APM v4.10.8 and v4.11.0rc7 both fixed a memory leak in database monitoring that grew with query volume — a critical patch for anyone running DBM propagation.

Sentry detects noise and enables equations

Sentry shipped two new features that change how you interact with data. The low-value span detection (beta) automatically surfaces high-volume, low-signal spans as a new issue type, complete with projected 30-day volume, estimated storage cost, and an auto-fix button. If you've ever wondered which spans are burning your quota without telling you anything useful, this is the answer. Separately, Sentry Dashboards now support equations for the Application Metrics dataset, letting you combine metrics into formulas like error percentages or custom Apdex scores — and you can save them directly from Explore.

Bug fixes and platform updates

Across the Datadog Python APM line, v4.8.12, v4.9.5, and v4.10.8 all fix a rare hang after child process forking. The v4.10.7 release also fixes that same database monitoring memory leak, and the Data Streams Monitoring SQS bug where JSON payloads broke pathway context extraction. On the browser side, Datadog RUM v7.5.0 now attaches debug IDs from source code context to error and long task events, and adds a beta enableViewUpdates option. Sentry's JavaScript SDK 10.64.0 introduces a dedicated nodejs_compat entrypoint for Cloudflare Workers, letting you use Node SDK features like prismaIntegration and vercelAiIntegration on Workers, while also reducing overhead by defaulting to Sentry's minimal SentryTracerProvider instead of the full OpenTelemetry SDK pipeline.