Has Grafana Taken Lead In Ai For Observability?

Grafana Labs has integrated AI support into its observability platform and panel, offering applicable improvements to nan observability experience.

Whether Grafana is surging up of nan observability vendor battalion aliases not depends connected really 1 measures nan different vendors’ offerings.

Providers presently clasp diverging philosophies connected really champion to usage AI for observability. DataDog’s AI is very overmuch limited and built connected Model Context Protocol (MCP) agents. Kloudfuse’s AI functionality is mostly successful support of information reservoir observability. Chronosphere is not arsenic bullish connected AI’s worthy and applicability to support observability astatine this time.

Meanwhile, I tin attest firsthand that Grafana’s AI support is useful erstwhile mounting up and integrating Grafana pinch OpenTelemetry, bringing metrics information from some Amazon Web Services and Google Cloud Platform information sources to stitchery metrics to my Grafana panel.

At a shop held astatine October’s ObservabilityCon successful London, I recovered myself stuck while creating a panel. I simply asked nan AI adjunct to group up a sheet for me, which it completed successful conscionable a fewer seconds. When I sewage stuck astatine different points during nan shop — specifically while mounting up nan sheet to merge AWS information into GCP — it offered maine applicable ways to proceed (see image below).

My acquisition was constricted during nan workshop, of course, and ReveCom, my expert firm, has yet to put it done nan paces to much profoundly analyse Grafana’s AI effort. However, I tin affirm that it was useful. That says a lot, and it was decidedly much applicable than conscionable asking questions done Cursor aliases Claude adjunct — aliases worse yet, pinch ChatGPT, which is often much worthless than not.

Automating Rollbacks

During ObservabilityCon, Grafana described Grafana Assistant arsenic an AI-powered chat integration built straight into Grafana, enabling users to interact pinch observability information done earthy language. It uses large connection models to make queries, analyse results and iterate intelligently.

The Assistant is designed to thief some non-technical users — those who mightiness inquire questions without building dashboards — and master tract reliability engineers who tin customize behaviour utilizing rules and infrastructure context. It connects pinch devices for illustration GitHub, AWS, and ticketing systems done MCP servers, and features “infrastructure memory,” which maps telemetry to understand dependencies.

Overall, Grafana Assistant was designed to make observability much accessible for method and non-technical folks. It accelerates investigations and integrates AI seamlessly into workflows, according to a number of speakers during ObservabilityCon.

Integrations pinch MCP servers, semantic connection rules for metrics and traces, and nan expertise to output diagrams are provided.

Additionally, Grafana’s Assistant Investigations, an autonomous supplier strategy for troubleshooting, was introduced astatine ObservabilityCon. This characteristic uses information sources for illustration nan Extended Berkeley Packet Filter (eBPF) and runs successful nan inheritance to research issues from aggregate angles.

.@grafana Assistant (an in-console AI co-pilot for observability) is now mostly available. At #ObservabilityCon today, Dmitry Filimonov demoed Assistant Investigations that lets nan supplier observe issues tin nan inheritance and tin thief analyse nan sources of errors. pic.twitter.com/xtrfw39wwT

— BC Gain (@bcamerongain) October 8, 2025

Grafana Labs CTO Tom Wilkie described it to maine this measurement during nan conference: “The conception of AI assistance focuses connected making AI really useful now, not conscionable a early promise. The extremity is to bring existent worth now by making it easier for customers to get started and diagnose problems. The soul accuracy is to use AI to trim nan toil for on-call engineers.”

The AI assistance features, specified arsenic nan adjunct and swarming investigations, urge adjacent steps (e.g., expanding relationship scaling). Internally, Wilkie said, nan institution is already automating nan exertion of these fixes.

A cardinal action automated is nan rollback of releases: Engineers are trained to rotation backmost releases instantly alternatively than spending 30 to 40 minutes applying a risky fix, but quality quality often causes them to effort to resoluteness nan problem properly, Wilkie said.

“Automating nan rollback achieves nan desired behaviour and handles a batch of nan toil for our engineers,” he said. “The imagination is that successful a fewer years, 70 to 80% of on-call load will conscionable beryllium handled autonomously by safe actions for illustration rollbacks and scaling.”

This attack is imaginable because nan package is built connected abstractions for illustration service-level objectives (SLOs), which let agents to understand if a merchandise is simply a regression aliases not without knowing nan ins and outs of each service, Wilkie said. Grafana’s heavy grounding successful unfastened root has played a domiciled arsenic well.

As he noted, “The institution believes its unfastened root instauration is its superpower erstwhile it comes to AI, arsenic nan models are already trained connected nan contented generated by their unfastened root projects.”

Group Created pinch Sketch.