Grafana Labs

📊 Start with a question in Claude. Keep exploring in Grafana. We’re proud to be a launch partner for Claude Dashboards, now in beta! Learn more: https://lnkd.in/e4ZH7BaB

The 5th Annual Observability Survey by Grafana Labs is now open, and we want your anonymous input! 🫵 It only takes 5-10 minutes and your input will help shape the upcoming report. Participants who complete the survey will have a chance to win a Grafana hoodie! Two winners will be announced every month from Oct-Dec 2026. Take the survey: https://lnkd.in/gi7zpKQj

🔎 Conveo, an an AI-led market research platform, adopted Grafana Cloud to unify metrics, logs, and traces through OpenTelemetry... and Grafana Assistant changed how engineers interact with the telemetry. Engineers who have never written a query language can now investigate incidents and build dashboards on their own. "For me, good observability means: 'Can you ask any question about the system and get a clear, actionable answer?'" — Ludovic Vannoorenberghe Senior Platform Engineer, Conveo Read their case study: https://lnkd.in/ggiRGkum

🤖 Running an LLM as your agent judge works... until your agent outgrows it. We rebuilt its evaluator from scratch (agreement with a stronger reviewer went from 2.1 to 59.5 out of 100), then immediately had to ask: does a typed-decision model like Jev make it obsolete? Short answer: not quite. Jev costs less than a third of the generative judge and runs more than 35x faster per call. But it agrees with the reference reviewer about half as well — and how you frame the decision matters more than you'd expect. Splitting evaluation into three narrow checks beat a single ranked scale by ~10 points. 👋 Yasir E. Ekinci and Sven Großmann walk through the full experiment: what broke in the original judge as agent sessions grew longer and more complex, what it took to fix it, and where Jev actually fits — now built into Agent Observability as an opt-in judge provider for operational trace triage. https://lnkd.in/g3atvB8q

If you've ever wondered how Grafana thinks about AI for observability and observability for AI, this oldie-but-goodie podcast episode covers your questions. ICYMI, Mat Ryer spoke with Ship It Weekly about: 1️⃣ Why AI demos are easy, but production AI is harder 2️⃣ Why agents need observability, evals, and guardrails 3️⃣ Where LLM-as-judge patterns, traces, tool calls, and feedback fit 4️⃣ Why telemetry cost problems may repeat with AI workloads 5️⃣ Why UX matters when operators need to trust the answer 6️⃣Where AI can help SRE and platform teams today

💰 Platform reliability isn't just an engineering problem. For companies like Dabble, it's a commercial one. The Australian sports betting company moved from self-hosted Grafana to Grafana Cloud to give their SRE, Cloud, and Engineering teams a single observability view across metrics, logs, and traces as they expand into the US and UK. "Running our own stack meant our SRE team was spending time on infrastructure that wasn't theirs to own. Moving to Grafana Cloud gave us the managed foundation we needed, and Grafana Assistant has changed how our engineers actually work through an incident." - Andrei Goutnik With Adaptive Telemetry helping manage data costs and Grafana Assistant accelerating incident response during live events, this is observability tied directly to business outcomes.

🆕 Grafana Tempo 3.1 is here! Featuring Kafka improvements, new TraceQL metrics capabilities, and trace redaction built in. This release makes it easier to operate Tempo at scale and get more out of your tracing data. Get the scoop on 3.1: https://lnkd.in/gMrp--uv

We cut our frontend observability bill by 99%. Without deleting a single byte of data. At Property Finder, our fastest-growing observability cost wasn't logs or metrics. It was browser monitoring  and it was growing faster than our traffic. Real user monitoring billed per session makes sense at one scale. At ours, we needed a different architecture. The standard answer is sampling: keep 1 session in 10, drop the rest. Cheaper, sure. But you're paying for observability and deleting the observations. We refused. Instead, we changed the meter, not the data: Same telemetry Routed through a self-hosted Grafana Alloy receiver Into Loki and Tempo Billed per GB instead of per session The results: 70% off the bill in month one, while the migration was still ramping 99% off this month Same data, same dashboards, zero loss in visibility, and the part I like most: cost now scales with data volume, not raw session count. Fair warning, this isn't a free lunch. We now own the ingest layer, which means: restoring labels the hosted collector adds automatically, resolving geolocation at ingest, and hardening a public endpoint. Real engineering work. Worth it at our scale but you should know what you're signing up for. What made this possible is Otel - Grafana's stack is genuinely open. Alloy can receive it. Loki and Tempo can store it. We didn't work around Grafana, we went deeper into it. Huge thanks to the Grafana Labs team for working through the architecture with us, and to Braulio Barahona, Yasitha Bogamuwa, Alex Capalneanu, Taran Kambow, Luis Cappa, Nour Hasbini, Andrey Inikhov, and Dinuka Jayamaha for the collaboration, support, and engineering effort that made this possible. #observability Grafana Labs #Alloy #RUM #costoptimization Property Finder

🌌 #GrafanaEverywhere, even when chasing the Northern Lights. With the aurora particularly strong this weekend, hear how this Golden Grot Award-winning project by Mohamed Adem 𓂆 uses public data and Grafana Assistant to explore whether an aurora will actually be visible.

🆕 New Grafana Cloud IRM demo: declare an incident from Slack and Grafana Assistant goes to work automatically: https://lnkd.in/g4nmg3is In this walkthrough, Ashley Somerville and Edward Qian highlight how Grafana Cloud IRM enables you to: ∙ Ask Grafana Assistant directly in Slack; it already has context from the alert and thread, no extra info needed ∙ Declare an incident and auto-create a Slack channel, assign roles, and escalate, all without leaving the thread ∙ Watch the investigations agent build and test hypotheses using your logs, metrics, traces, and MCP integrations like GitHub ∙ Generate a post-incident report and upload it straight to Confluence, from the new Grafana mobile app