Ship It Conversations: Mat Ryer of Grafana Labs on AI Observability, Agents, Evals, and Operating AI in Production

This is a guest conversation episode of Ship It Weekly, separate from the weekly news recaps.

In this Ship It Conversations episode, I talk with Mat Ryer of Grafana Labs about AI observability, production agents, evals, telemetry cost, guardrails, and what changes once AI moves beyond demos and into systems teams actually depend on.

Mat is Senior Director of AI at Grafana Labs, where he focuses on how AI fits into observability and production systems.

We talk about Grafana Assistant, why AI observability is not just logs, latency, and HTTP 200s, and how teams can measure whether agents are actually helping. Mat gets into evals, LLM-as-judge patterns, traces as a way to think about conversations, user feedback, tool choice, model changes, and the cost of collecting new telemetry.

We also dig into UX and trust. If an AI assistant gives you a wall of text, you still have to decide whether to believe it. If it can show the graph, deep link into Grafana, apply filters, and expose the source data, that becomes a much more useful operating experience.

The big takeaway: start small, enhance workflows you already have, build feedback loops, and treat production AI like something you actually have to operate.

Highlights

• Why AI demos are easy, but production AI is harder

• Why agents need observability, evals, and guardrails

• How Grafana thinks about AI Assistant and AI observability

• Where LLM-as-judge patterns, traces, tool calls, and feedback fit

• Why telemetry cost problems may repeat with AI workloads

• Why UX matters when operators need to trust the answer

• Where AI can help SRE and platform teams today

Links

Mat Ryer on LinkedIn: https://www.linkedin.com/in/matryer/

Mat Ryer on GitHub: https://github.com/matryer

Grafana Labs: https://grafana.com

Grafana Assistant: https://grafana.com/products/cloud/ai-assistant/

Grafana AI Observability: https://grafana.com/docs/grafana-cloud/machine-learning/ai-observability/

Grafana Adaptive Telemetry: https://grafana.com/products/cloud/adaptive-telemetry/

Grafana MCP server: https://github.com/grafana/mcp-grafana

OpenTelemetry: https://opentelemetry.io

Prometheus: https://prometheus.io

Grafana Loki: https://grafana.com/docs/loki/latest/

More episodes and show notes: https://shipitweekly.fm

On Call Brief: https://oncallbrief.com