Grafana Labs Targets AI Blind Spot

To drive this new approach, Grafana Labs has created a new AI organisation under Mat Ryer, Senior Director of AI at Grafana. Ryer has been with Grafana since 2021, but this new position will allow him to create and develop his own team.

On the main stage, Ryer said, “We want to be the first AI native platform, and that means dealing with non-deterministic inputs. We’re dedicated to building safe, secure, actually useful AI, and we take this quite seriously.

“We don’t want to ship slop. We don’t want to be a Slop Shop. So everything we deliver, everything we’re doing here, is about guaranteeing when we ship something to you, it is going to actually make a big difference in your day-to-day work.”

Grafana Labs is not alone in this assessment. Precisely has been making this point about observability over the last couple of years. Research conducted with Drexel University, LeBow College of Business, shows that AI observability and AI maturity are major issues for many organisations.

If You Can’t Observe It, You Can’t Control It

There are several parts to the Grafana solution. At the core is a new agent interface. It embeds Grafana Cloud’s full observability capabilities directly into AI-assisted environments like Cursor, Claude Code, and GitHub Copilot. This integration reduces context-switching between code and production. That allows developers to monitor agent behaviour, including inputs, outputs, and execution flows, in real time.

By allowing developers, operations and security teams to observe AI, they are in a better position to manage it. The new solution continuously evaluates outputs. Continuous monitoring provides a real-time service.

It surfaces risks such as data exposure or anomalous usage patterns, and elevates agent sessions alongside existing application telemetry. This is about debugging, governance, compliance, and performance at scale.

Importantly, the platform correlates agent behaviour with first-class telemetry signals. It allows teams to detect low-quality responses, policy violations, or behavioural drift before they impact users. For any organisation deploying AI agents in customer-facing or regulated environments, this level of visibility is essential.

Expanding Grafana Assistant Beyond the Cloud

Grafana Assistant, the company’s AI-powered observability agent, is also expanding beyond Grafana Cloud. It now supports on-premises deployments of Grafana Enterprise. It addresses a need for enterprises with strict data residency or control requirements. Importantly, it closes the AI observability gap and ensures that they meet compliance demands.

Ryer commented, “This is a sidebar chat app that’s integrated deeply into Grafana. It’s an LLM integrated into the very fabric of Grafana. It helps you do everything you need. You can just ask it questions. It knows how to write complex queries, so it can go and dig into the telemetry and give you the insights that you need. And it’s in a loop, so it can go around and gather relevant context, so it can make informed assessments.”

The Assistant now integrates with MS Teams, over 50 native data source integrations, and Python runtime. This makes it a central hub for automation, scheduled tasks, and remote operations. Grafana believes that this takes it beyond being an assistant and makes it the operational brain of the observability stack.

Features like “remote MCP server” let users bring their own agent and connect it to Grafana’s remote infrastructure. There is also a “learn mode” which offers personalised, hands-on lessons tailored to the user’s workflow. This gives users the ability to integrate AI as they develop new workflows. It takes away the need for IT to do it for them and gives them the observability.

A new Agent Interface for the Cloud

The Grafana Cloud CLI (GCX) is a new agent interface. It brings cloud observability directly into AI-assisted dev environments. That allows developers to query live observability insights. It gives them the ability to correlate alerts with recent code changes and propose fixes. All of this happens without them leaving the IDE.

This is a significant boost for DevOps and SecDevOps environments. It delivers continuous improvement driven by observability into that dev-production loop. If teams are using AI coding tools such as Cursor or Claude Code, GCX allows them to validate agent behaviour before deployment. The result is a lower risk of production incidents caused by hallucinations or misconfigurations.

o11y-bench: A Standard for Measuring AI Agent Effectiveness

Supporting this change to the DevOps environments is o11y-bench. It is an open-source benchmark designed to measure how agents perform real-world tasks. Few organisations have any clue how effective their AI agents are. Their focus, to date, has been on security. Now it is possible to query telemetry, investigate incidents, and modify dashboards.

o11y-bench also runs against the full Grafana stack. It evaluates what agents say and what they do. The latter will be interesting for many organisations, and it is likely that security teams will want to get involved here. They are seeing agents deployed across the enterprise, and tracking them is putting pressure on their tools. o11y-bench might just be the tool that spans DevSecOps for AI agents.

This also delivers a critical step toward standardisation. The proliferation of AI agents requires ways to compare performance, reliability, and safety across tools and vendors. o11y-bench is a common framework for assessing agent effectiveness in complex observability environments.

Enterprise Times: What does this mean?

Most of the focus on AI has been on how to get from pilot to production and scale. For many organisations, this is the wall they are hitting. Grafana Labs is taking a different approach. Rather than enter a crowded market of AI tooling, they are delivering observability. For overstretched security and operations tools, this is an area where they see immediate benefit.

Developers also gain here. Less context-switching, an ability to iterate faster and improved confidence in AI-generated outputs. The latter is critical to those who are looking at greater use of AI in software development. At the moment, they are not sure if the benefits outweigh the risks.

It also allows enterprises to leverage their existing Grafana investments to monitor AI agents. They don’t need a new platform or new tools that have to be integrated. It is right there inside Grafana, without needing to adopt entirely new platforms.

Moving to a world of AI observability that meets the needs of the whole business is a must, not an option. Grafana has chosen its path, and it will be interesting to see how customers respond.

The post Grafana Labs Targets AI Blind Spot appeared first on Enterprise Times.