Operations | Monitoring | ITSM | DevOps | Cloud

If you search for “OpenTelemetry Agent”, you will likely encounter two completely different definitions. This ambiguity often leads to confusion between infrastructure teams and application developers. SREs and DevOps engineers would describe it as a component deployed as a sidecar, whereas application developers would understand it as a language-specific library. Let’s break it down in the next section.

The air in the operations world is thick with AI and LLMs. EVERY vendor is rushing to slap an “AI-powered” badge on their product. But here’s the uncomfortable truth: In high-stakes Kubernetes operations, one bad AI recommendation can destroy months of trust-building in an instant. We aren’t building a chatbot to suggest recipes. We are building systems that, armed with kubectl permissions, have the potential to take down production with a single, wrong command.

2025 was the year we at Grafana Labs went all-in on AI—and boy, what a year it was. Not only did we establish and start to execute our overarching strategy (build actually useful AI), we also took one of our most exciting new features (Grafana Assistant) from idea to general availability in just nine months! Yes, there's no shortage of articles singing the praises of AI these days, but let's dispense with the hyperbole and focus on some actually useful content.

These days, engineering leaders find themselves caught between a rock and a hard place. On paper, AI adoption looks like an unqualified success. Developers are shipping more code faster than ever, pull request volumes are up, and teams report feeling more productive. Their leaders rush to LinkedIn to share their plans to scale adoption because their teams are just so much more efficient. But then, the incidents and bug reports start piling up.

Knowledge graphs and RAG (Retrieval-Augmented Generation) are complementary techniques for enhancing large language models with external knowledge, and each brings unique strengths for DevOps use cases. While they are often mentioned together, they are fundamentally different systems, and combining them delivers far better outcomes than relying on either approach alone.