Grafana Labs launches six AI tools for telemetry & agents

Grafana Labs has launched various new AI capabilities as part of the company's AI Week event, spanning new ways to interact with telemetry data, investigate incidents, and monitor the behaviour of AI agents in production, according to the company's Senior Director of AI, Mat Ryer.

Ryer said the event, which runs from July 27 to 31, was designed to showcase tools that had already proven useful to customers, rather than features built to capitalise on interest in AI.

"We were a little bit late to the party, but that gave us a good position to see what was out there, what was working and what wasn't. A lot of things out there are sort of imagined, and you can tell. We're not interested in AI for the hype or for the sake of it, so we really wanted to make sure we had something concrete, something real, something that actually moved the needle for customers before we released things," said Ryer.

Earlier this week, Grafana Labs made its Workspace tool generally available, an interface built around a conversational agent rather than manually built dashboards.

"This is an agent-first immersive experience where you start with the agent. You start with a conversation, something you're interested in or concerned about, or a question you have for your telemetry. You ask this in natural language, and the agents go and answer that question," Ryer said.

The company made its command-line tool, gcx, along with its cloud and open-source MCP server, generally available as well, having previously offered earlier, more limited versions of the capability. Ryer said the tools are designed to bring Grafana's agentic capabilities directly into coding agents, so that a planned code change can be informed by real production telemetry rather than developed in isolation from it. Reaching general availability means the underlying API is now fixed, with a backwards-compatibility commitment that was not previously in place.

Grafana Labs' Assistant Investigations feature is generally available, alongside Assistant Automations, and moved a third capability, Watcher agents, into public preview.

Investigations dispatches multiple specialist agents to examine an alert by following leads and testing possible explanations before a person looks at it.

"It comes up with a list of hypotheses and sets about disproving them. There's an adversarial element to it, which I think is quite interesting - it will challenge itself internally, and that's how we're able to have a higher level of confidence in what it's concluding," Ryer said.

Automations let teams schedule natural-language reports, such as a weekly summary of incident themes, delivered directly into Slack or Microsoft Teams, according to Ryer. Watcher agents, meanwhile, are designed to continuously monitor situations that are harder to define with a fixed metric or threshold, taking action autonomously once a user has tested and confirmed its behaviour.

On Thursday, Grafana Labs made its Agent Observability capability generally available, giving teams visibility not only into an AI agent's latency and cost, but into the substance of its conversations, including whether it gave sound advice and used the right tools. 

Tomorrow, the company is extending its Grafana Assistant into Slack, Microsoft Teams, and mobile and desktop apps to finish off the week.

AI Week was timed to fill the gap between Grafana Labs' two annual conferences, GrafanaCon and October's ObservabilityCON in San Francisco, given how quickly the AI and observability space is moving. He said much of the customer engagement around the launches was happening through community channels such as Reddit and podcast recordings, with the company encouraging customers to test the tools against their own production data rather than rely on demonstrations.

Ryer said the aim throughout is to let teams verify that an agent's output can be trusted rather than take it on faith.

"That helps you really trust that these agents are actually doing what you want them to, and that you can believe the results."