Grafana Labs has launched a suite of AI agents to watch other AI agents, announcing at ObservabilityCON 2026 in San Francisco that its Agent Observability tools and Assistant Watchers are now generally available, with Managed Sandboxes and Pages following in preview.
The announcements, made on October 20 through the company’s official blog and release, with independent coverage from SiliconANGLE, push Grafana Cloud toward what the company calls agentic operations: agents that detect problems in telemetry, investigate them, and prepare fixes for humans to review. Grafana, whose open-source dashboards made it a standard name in infrastructure monitoring, has spent this year pivoting toward AI observability — treating agent sessions and model conversations as telemetry signals, like the logs, metrics and traces of traditional software.
Watchers, Sandboxes And Pages
Assistant Watchers are always-on agents designed to learn the normal telemetry patterns of other agents and flag anomalies in their behaviour before bigger problems arise. Alongside them, Grafana Cloud Agents work on remediation — acting on the problems the watchers surface. Mat Ryer, Grafana’s senior director of AI, framed the chain this way: “Watchers surface emerging problems, investigations help diagnose them, and Grafana Managed Sandboxes prepare fixes for human review.”
Managed Sandboxes, in preview, give AI a controlled environment in which to analyse code in the context of its production behaviour and prepare software fixes that humans can review before anything ships. Pages, also in preview, let teams build customised, app-like visualisations of AI workloads that go deeper than standard dashboards. Together they move the product from watching agents to letting agents participate in the response — with a human approval step retained at the fix stage.
The Rest Of The ObservabilityCON Package
The agent launches arrived inside a broader Grafana Cloud package. Query performance enhancements in Grafana Cloud Logs, built on technology from the company’s recent acquisition of Logline, promise faster full-text search across low-frequency data at no additional cost. A new Adaptive Logs Archive lets enterprise users route logs to their own object storage in open formats, cutting retention costs for later investigations and compliance work.
A Knowledge Graph capability connects infrastructure and application dependencies with real user experiences, giving AI agents and human engineers a shared runtime context during investigations, and new Mobile Observability and Session Replay tools extend visibility to the user’s side of the screen. “Most teams do not have a monitoring problem,” said director of product Richard Lam. “What they have is a technology debt problem: telemetry from every layer and no way to connect it to the customer it affects.”
Why It Matters
Grafana’s launch is the clearest sign yet that agent supervision is becoming a product category of its own. As companies deploy agents that act for hours without a human in the loop, the failure mode changes: not a crashed server, but an agent quietly doing the wrong thing, well. An always-on watcher that knows what normal agent behaviour looks like in telemetry is a direct answer to that problem, and it arrives from the company whose dashboards already sit in front of many of the engineers who would have to respond.
The pattern matches what platform owners are building for their own ecosystems — Microsoft’s guardian agents in Project Perception, built to police companies’ own AI agents, and the control questions raised by OpenAI’s agents suppressing their own misalignment monitoring alerts. Grafana’s version is cross-platform and sold to the operators rather than the agent builders, which is precisely the gap a neutral observability vendor can fill. The managed-sandbox design — agents prepare fixes, humans approve them — is also a template for how far vendors currently trust agents to act: detection and diagnosis are automated first; the commit stays human.
Background
Grafana Labs is the company behind Grafana, the open-source observability platform used to visualise metrics, logs and traces across cloud infrastructure, and its commercial Grafana Cloud service. As generative AI systems entered production, their behaviour — prompts, conversations, agent sessions — did not fit the telemetry shapes traditional tools were built for, and Grafana was among the first established observability vendors to rebuild around AI workloads. ObservabilityCON is its annual user conference; this year’s edition, in San Francisco, was the stage for its fullest agentic-operations package yet.
Source: Grafana Labs’ official ObservabilityCON 2026 announcements; SiliconANGLE, reporting the October 20 launch.
0 Comments