Grafana ObservabilityCON 2026: Cheaper Telemetry, More AI Agents

Grafana Labs says customers of its Adaptive Telemetry suite have cut telemetry spend by 30% to 50% on average, and at ObservabilityCON 2026 it pushed that cost story further while handing more of incident work to AI agents. For platform teams paying for observability by volume, the useful part of the announcements is not the agents but the new ways to drop data without losing the one trace or log line an investigation needs.

What Happened

Grafana’s ObservabilityCON 2026 announcement recap groups the news into three themes: autonomous operations, cloud economics and context.

On the AI side, Assistant Watchers are now generally available. They are always-on agents that watch for changes predefined alert rules miss: a team describes what to monitor in plain language, picks data sources, and the watcher builds a baseline from historical telemetry and flags shifts. Grafana Managed Sandboxes, in public preview, let Grafana Assistant use repository code and live telemetry to propose fixes and prepare pull requests in a sandbox, which a person on the team then reviews and approves. Grafana Pages, also in preview, builds app-like visualizations from a description.

Why It Matters

The cost features are where the money is. Grafana’s Adaptive Telemetry recommends what to aggregate or sample across metrics, logs, traces and profiles, and the updates close the gaps that make teams hesitate to cut:

• Adaptive Metrics auto-apply (generally available) keeps applying recommendations without breaking dashboards and alerts. Cost-first mode only applies changes that won’t raise total active series.
• Adaptive Logs Archive (public preview) routes logs to an object storage bucket the customer owns, with on-demand replay in private preview.
• Querying traces dropped by Adaptive Traces (generally available) retrieves any trace by ID from the last 24 hours, even if it was sampled out.
• Volumetric sampling (generally available) spreads the sampling budget across services and request types, and in early testing delivered about 25% higher information density than probabilistic sampling at the same volume.

Full-text search in Grafana Cloud Logs, built on technology from Grafana’s April 2026 Logline acquisition, is now generally available at no extra cost. DATEV’s Timo Kirzdörfer reports going from processing nearly a terabyte of logs in three minutes to around 100 gigabytes in under three seconds for newly submitted logs.

What This Means for Platform Teams

The context work ties the two halves together. Grafana is making its Knowledge Graph, the map of components, relationships and health, a capability in every stack, and adding frontend apps, more AWS, Azure and Google Cloud resources, and, in preview, AI agents and SQL Server databases. It is also available as runtime context for AI agents, including a team’s own agents through gcx. The more an agent can see, the more useful Watchers and sandboxed fixes become, and the more a team needs a review step before anything ships.

Stack Archive’s guide to Adaptive Metrics cost-first versus coverage-first auto-apply covers the trade-off behind the first of these cost features in detail.

Source: Grafana Labs Blog

Content created with AI assistance and reviewed for accuracy.