Grafana Labs has announced new products and updates for Grafana Cloud focused on AI agents and monitoring AI systems. The releases include the general availability of Agent Observability with OpenTelemetry support, as well as Assistant Watchers.
The move comes as more IT and engineering teams put AI tools into pilots and production, according to survey data the company released.
Among the new additions, Assistant Watchers are always-on agents that look for changes in telemetry data not covered by pre-set alert rules. Users describe what matters in plain language, and the tool learns normal patterns before flagging changes through Grafana Alerting.
Critical findings from Watchers can also trigger an Investigation. The feature is intended to sit alongside threshold-based alerts and help teams spot emerging problems.
Grafana also introduced Grafana Managed Sandboxes in public preview. The feature gives Grafana Assistant a controlled cloud environment where it can work with repository code and live telemetry, investigate code in the context of production behaviour, prepare a fix, and open a pull request for human review.
Another preview product, Grafana Pages, is aimed at teams that want to build visualisations with fewer layout constraints than a standard dashboard. Existing visualisation components will continue to work within the new format.
AI operations
The announcements expand Grafana's work on what it describes as agentic operations, in which software agents take on more operational tasks while human teams retain oversight. Grafana Assistant is already embedded in Grafana Cloud, allowing users to query telemetry in natural language, while Assistant Investigations can run longer automated investigations triggered by alerts or incidents.
"Teams want agents to take on more of the work, but they need evidence and control at each step," said Mat Ryer, Senior Director of AI, Grafana Labs. "Watchers surface emerging problems, Investigations help diagnose them, and Managed Sandboxes prepare fixes for human review. Together they make Grafana Cloud more autonomous, bringing detection, diagnosis, and proposed action into one workflow, grounded in the team's own telemetry."
Survey findings cited by Grafana suggest adoption is moving ahead, but comfort levels vary by use case. In a survey of 300 IT and engineering leaders, 68% said they had moved AI for observability beyond investigation into pilots or production. Meanwhile, 85% said they were comfortable with read-only investigation, while 52% were comfortable with full autonomous remediation.
The same survey found that 80% of organisations are building or operating large language model-based features. Yet 45% said they had assembled AI observability from open-source tools, while 25% used a purpose-built product.
Watching the agents
Grafana's second area of focus is monitoring the behaviour of AI agents once they are in production. Agent Observability, now generally available with OpenTelemetry support, is intended to show what agents are doing, what they cost, and how their behaviour changes over time, using the same standard as the rest of an organisation's systems.
With OpenTelemetry support, agent activity can be tracked in the same way as other services in the software stack. That matters because many organisations already use the open standard across their observability tools.
Grafana also pointed to separate IDC research on users of its AI tools. According to the study, customers using Grafana Cloud's AI tools reduced mean time to resolution by 63% and spent 46% less of each on-call shift gathering context.
One customer, identity security company SailPoint, said it has used Grafana Assistant for telemetry migration work and incident investigations.
"I can just ask Grafana a question, leave it in that tab, go do something else, get Slack, get some coffee, and come back to a full investigation," said Juan Carlos Ramirez, Senior Staff DevOps Engineer, SailPoint. "It's personally helped me find spikes in logs that before the Assistant that would've taken me maybe 15, 20 minutes down to maybe five, and that's while I'm doing other things."