Instrument and observe production-ready agentic apps with Grafana Cloud and AWS

GenAI applications do far more than chat: They reason, call tools, retrieve knowledge, and take actions inside business-critical workflows. This all creates new operational challenges in understanding whether your agents are behaving correctly, safely, efficiently, and cost effectively in production.

In this workshop, you'll see how to build an agent using Amazon Bedrock and Amazon Bedrock AgentCore, which are AWS services designed to help teams build, deploy, and operate agents securely at scale. From there, you’ll use Grafana Cloud's Agent Observability to monitor a real multi-agent AI application and add the observability, evaluation, alerting, and guardrail layers needed to run the agent with confidence.

Through a series of guided labs, you’ll inspect conversations, prompts, tool calls, and agent execution to understand why an agent behaved the way it did. You'll create evaluators to automatically detect prompt injection, toxicity, and bias, configure evaluation rules, create alerts when AI behavior degrades, and finally deploy guards that block malicious requests before they ever reach the model.

By the end of the session, you’ll understand how to move beyond traditional application monitoring and build a complete operational layer for AI systems. You’ll leave with practical patterns for optimizing cost and performance, debugging agent behavior, protecting users, and continuously improving AI applications in production.

Whether you're building copilots, autonomous agents, or customer-facing AI applications, you'll leave with practical techniques you can immediately use to observe, evaluate, govern, and improve AI systems running in production.