Grafana cofounder says community looking for flexibility, not demanding AI...yet
“Our approach right now is to sort of look at how AI can elevate a junior SRE to a more experienced SRE..."
After years of helping of helping data specialists and DevOps types create “beautiful dashboards”, Grafana’s cofounder Torkel Ödegaard is now contemplating how to help enterprise customers make better sense of the underlying data as well appreciate the pretty visualisations.
That means working out how the open source observability platform can reduce the need for queries and allow customers – especially in the enterprise – to simply dive through data without getting bogged down constructing queries.
It also means working out where artificial intelligence fits in, without trumpeting the technology in “a dishonest way”.
Speaking to The Stack at the vendor’s GrafanaCon event in Amsterdam recently, Ödegaard explained the dynamic between the open source Grafana project and the commercial organization Grafana Labs.
The vendor’s cloud service aims to be a “more out of the box experience”, he said, but while the open source project strives for ease of use.
“It's also this very flexible toolbox that anyone can build an observability solution around by plugging in different data sources, building dashboards, really kind of owning and building that their solution.”
But, he says, increasingly the aim is how to have Grafana “assemble the solution itself” guided by the user and being smarter about the underlying data. “I want to go to a place where we don't have to build so many custom dashboard…that Grafana can automatically maybe do some of that for you dynamically.”
This is helped by the fact there is more metadata about what underlying metrics mean, as well as more standardization, for example around OpenTelemetry, he said.
Grafana Labs unveiled Explore Metrics at the event, allowing users to drill down through metrics without getting bogged down in generating queries. It also unwrapped a similar project aimed at logs, which had been developed just weeks before at a company hackathon.
To date, Grafana has been agnostic about what data actually means, leaving inference to the user. “What I think we should be able to do is also create links between different related data …. where we can leverage the metrics in a smart way that the data and the metadata associated with the metrics to come up with associations between things and also have some understanding of what these things actually mean represent.”
See also: How Japan's space agency used dashboards in its race to the moon
While the metrics and logs tools are open source, he said he envisioned proprietary or cloud only addons to make UIs richer. “I think is going to just reduce the need for many dashboards because you can maybe often suffice with an Overview Dashboard, and then you can use these automatic drill-down views to, to look at more fine-grained [information and data].”
This begs the obvious question whether AI will play an increasing role in Grafana’s offering. Ödegaard said, “I think there's definitely more intelligence in these auto query UIs, where we actually try to come up with the best suitable query for a metric.” Deducing the relationships between metrics and connections between elements could involve a form of machine learning, he continued.
It remains to be seen what other ways AI could play a part in understanding the connections and their meaning, he said. “I think we'll definitely try to use one of the open models and then train... them,” he said, to produce recommendations and act as a copilot. “Our approach right now is to sort of look at how AI can elevate a junior SRE to a more experienced SRE.”
Or as Grafana CTO Tom Wilkie told us separately, tools such as LLMs will “augment the experience”. In the case of junior engineers, he said, “because the LLM will make sure you're asking the right questions, will help you in your root cause analysis. I don't think it will automate them away though.”
At the same time, said Ödegaard, Grafana has to be wary of clambering aboard the bandwagon and say “we have AI too” in a dishonest way. Companies of all sorts face a challenge in “not being excluded” from the AI race, while actually delivering “real value.”
Interestingly, for now at least, the Grafana community is not clamouring for AI features he said. “The most requests are really around more flexibility and more options.” These were perennial challenges as a product designer, or manager, he said, “We do want to be very powerful, flexible tool that can cancel, support moon landings, and do these really weird cool things,” he said, while also being ever more intuitive and easy to use – characteristics that are more important as Grafana is adopted by enterprises.
More broadly, the last year has seen upheaval in the open source world, whether it’s HashiCorp changing its license conditions and sparking the creation of OpenTofu in turn, or the CNCF having to step to preserve the FluxCD projects after its main backer Weaveworks was forced to close its commercial operations.
“We feel like almost last company which is trying really doing our best to stay true to our roots of being open source and having an open source licence,” Ödegaard said.
Most contributions to open source Grafana come from the company’s engineers, but Ödegaard said this was in large part because the company had taken on most of the key contributors. Nevertheless, he said, the broader community was essential, whether in terms of feature requests, fixing bugs or working on pet use cases. As for Ödegaard, “I'm still very active as a developer and UX designer."
