From Diagnosis to Action: Building AI Agents That Fix Problems

The goal of observability isn't only understanding what's broken, it's fixing it.

This session demonstrates how to build agents that reason through uncertainty, explain their confidence, match symptoms to remediation playbooks, and then propose fixes.
Using Elastic's Agent Builder, as well as a simple Go-based environment, we will construct mock microservices that can be intentionally degraded, along with remediation playbooks stored in Elasticsearch with semantic search. Agent Builder orchestrates the AI reasoning by using ES|QL tools to parse and search relevant logs and metrics, and by using index searches to find relevant playbooks, even when the symptoms don't match exactly.

We'll go over best practices for adding human-in-the-loop components that let your agent run read-only operations autonomously, while requiring human confirmation for destructive actions.

Lernziele

Attendees will learn how to leverage their logs and metrics with AI Agents that know how, and when, to solve problems, showing just how powerful observability can become!