AI agents and smarter grouping reshape the observability landscape

Smarter error grouping at Sentry

Sentry upgraded the AI model behind its issue grouping engine, replacing off-the-shelf code embeddings with a model trained specifically on real Sentry stack traces. The result is a 50% reduction in incorrect merges and a 20% drop in duplicate issues. The new model is calibrated to err on the side of separation, meaning distinct bugs are far less likely to be lumped together — a meaningful change for teams relying on issue feeds to prioritize work. Sentry also shipped a major update to Sentry's JavaScript SDK, deprecating sendDefaultPii in favor of a new dataCollection option that gives fine-grained control over each category of data collected, while adding official support for Angular 22.

Datadog's AI agent and product wave

Datadog had an unusually dense week, launching more than a dozen new capabilities centered on AI-driven workflows and data interoperability. The standout is Bits Release, a product that helps teams ship code safely at the speed of AI-generated PRs — now accounting for tens of millions of commits per month. Bits Investigation (rel:rel_Yhc1GXXdkTG5zCtdf7Hrr) and Bits Testing (rel:rel_TO2uN7cnHTD1HS16bvTh1) round out the suite, respectively automating root cause analysis for synthetic test failures and generating adaptive test coverage from user journeys.

For AI agent observability, Datadog introduced native integrations with Claude Code, Claude.ai, Cursor, ChatGPT, and Codex via its MCP Server, plus MCP Apps for building interactive experiences inside AI workflows. The Agent Console lets teams monitor coding agent adoption, spend, and waste patterns.

On the infrastructure side, the ClickHouse integration partnership (in Preview) lets teams route high-volume logs to ClickHouse via Observability Pipelines and search them from the Log Explorer without re-ingestion. The Federated Logs capability takes this further, enabling investigations across Datadog, Databricks, ClickHouse, and SIEM tools from a single query surface. For security teams, Code Threat Detection monitors PRs for supply chain attacks, while Cloud SIEM automates threat hunting.

Dash0 smooths the dev loop and adaptive refresh

Dash0 overhauled its CLI experience with browser-based OAuth login via PKCE, a local OTLP forwarder for in-loop development, precision-mode queries, and one-command Homebrew installs. The adaptive auto-refresh feature replaces the old fixed-interval picker with a single toggle that matches refresh cadence to chart resolution, pauses polling when you scroll into data, and stops entirely when the tab is backgrounded — small ergonomic upgrades that add up during long debugging sessions.

Datadog Agent and APM: lazy trace agents, allocation profiling

Datadog Agent 7.80.0 makes the APM trace agent start lazily on Linux (no listeners until data arrives), reducing baseline resource consumption. The release also adds NCCL metrics for GPU workloads and support for OTLP delta sum metrics with rate conversion. Datadog APM v5.108.0 adds allocations profiling support and fixes mark-sweep GC event handling. Multiple APM patch releases addressed cgroup v2 container tag gaps, IAST/profiling crashes, and data loss for LLM observability on specific Datadog sites.

Keeping your service map straight

Service Remapping helps teams maintain consistent service naming across APM, Log Management, and other Datadog products — a small but important fix for the kind of drift that silently breaks correlations.