Six improvements that make Grafana Assistant better at helping your workflows

Grafana Labs reports six months of improvements to Grafana Assistant aimed at making dashboard creation and incident investigation faster, more accurate, and easier to verify through evidence. Dashboarding changes, guided by production telemetry and reproducible evaluations, reportedly reduced the error rate from 2.47% to 0.65%, lowered 95th-percentile operation latency from 63 to 28.8 seconds, and cut request costs by 18%, while the test suite expanded substantially. The company emphasizes evaluation loops that assess not only response text but also queries, tool calls, saved dashboard state, and conclusions, allowing teams to test changes to prompts, tools, and models against consistent scenarios. Other efforts include gcx, an agent-oriented CLI with self-describing commands and recovery-focused errors; progressive tool discovery, which reduced measured initial tool-definition context from about 54,500 tokens to 504; and improved search that represents dashboards and incidents in more meaningful forms while enforcing current permissions. Testing investigation agents against real resolved incidents increased performance on an internal difficult-incident benchmark from 52% to 69% in one week, while evaluator experiments found that a faster, cheaper structured-decision model was less aligned with a stronger generative reviewer for primary quality scoring. Together, these projects illustrate Grafana Labs’ iterative approach to AI development, using customer problems, evaluations, traces, and controlled experimentation to improve Assistant and Agent Observability capabilities.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.