What you'll learn
- How to trace multi-step agent workflows (e.g. classify → enrich → respond) and see tool and knowledge-base calls in context
- How to attribute token cost and latency by provider and agent, and find optimizations (e.g. caching)
- How to run online evaluations (PII, prompt injection, quality) and use production-backed recommendations to improve prompts and agent versions
See conversations, traces, token cost, and live quality checks for AI agents—correlated with your apps and infrastructure in Grafana Cloud.
Your support and platform teams are shipping agentic workflows—but tickets, CSAT, and LLM spend can move in the wrong direction fast. In this session, you’ll see how Grafana AI Observability unifies behavior (conversations and agent graphs), cost (tokens and providers), and quality/safety (online evals) with the metrics, logs, and traces you already collect in Grafana Cloud. We’ll walk through a real multi-agent support scenario: controlling cost across providers, catching PII and prompt-injection attempts, tracing failures across agents and tools, and improving prompts using version scores and AI recommendations—with Grafana Assistant to go deeper when you need it.
Your guide
Abdelkrim Hadjidj
Solutions Engineering Manager | South & Central EMEA
Grafana Labs